2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
42 #include <linux/btrfs.h>
43 #include <linux/blkdev.h>
44 #include <linux/posix_acl_xattr.h>
45 #include <linux/uio.h>
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
55 #include "compression.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
64 struct btrfs_iget_args
{
65 struct btrfs_key
*location
;
66 struct btrfs_root
*root
;
69 struct btrfs_dio_data
{
70 u64 outstanding_extents
;
72 u64 unsubmitted_oe_range_start
;
73 u64 unsubmitted_oe_range_end
;
76 static const struct inode_operations btrfs_dir_inode_operations
;
77 static const struct inode_operations btrfs_symlink_inode_operations
;
78 static const struct inode_operations btrfs_dir_ro_inode_operations
;
79 static const struct inode_operations btrfs_special_inode_operations
;
80 static const struct inode_operations btrfs_file_inode_operations
;
81 static const struct address_space_operations btrfs_aops
;
82 static const struct address_space_operations btrfs_symlink_aops
;
83 static const struct file_operations btrfs_dir_file_operations
;
84 static const struct extent_io_ops btrfs_extent_io_ops
;
86 static struct kmem_cache
*btrfs_inode_cachep
;
87 struct kmem_cache
*btrfs_trans_handle_cachep
;
88 struct kmem_cache
*btrfs_transaction_cachep
;
89 struct kmem_cache
*btrfs_path_cachep
;
90 struct kmem_cache
*btrfs_free_space_cachep
;
93 static const unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
94 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
95 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
96 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
97 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
98 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
99 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
100 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
103 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
);
104 static int btrfs_truncate(struct inode
*inode
);
105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
);
106 static noinline
int cow_file_range(struct inode
*inode
,
107 struct page
*locked_page
,
108 u64 start
, u64 end
, int *page_started
,
109 unsigned long *nr_written
, int unlock
);
110 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
111 u64 len
, u64 orig_start
,
112 u64 block_start
, u64 block_len
,
113 u64 orig_block_len
, u64 ram_bytes
,
116 static int btrfs_dirty_inode(struct inode
*inode
);
118 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
119 void btrfs_test_inode_set_ops(struct inode
*inode
)
121 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
125 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
126 struct inode
*inode
, struct inode
*dir
,
127 const struct qstr
*qstr
)
131 err
= btrfs_init_acl(trans
, inode
, dir
);
133 err
= btrfs_xattr_security_init(trans
, inode
, dir
, qstr
);
138 * this does all the hard work for inserting an inline extent into
139 * the btree. The caller should have done a btrfs_drop_extents so that
140 * no overlapping inline items exist in the btree
142 static int insert_inline_extent(struct btrfs_trans_handle
*trans
,
143 struct btrfs_path
*path
, int extent_inserted
,
144 struct btrfs_root
*root
, struct inode
*inode
,
145 u64 start
, size_t size
, size_t compressed_size
,
147 struct page
**compressed_pages
)
149 struct extent_buffer
*leaf
;
150 struct page
*page
= NULL
;
153 struct btrfs_file_extent_item
*ei
;
156 size_t cur_size
= size
;
157 unsigned long offset
;
159 if (compressed_size
&& compressed_pages
)
160 cur_size
= compressed_size
;
162 inode_add_bytes(inode
, size
);
164 if (!extent_inserted
) {
165 struct btrfs_key key
;
168 key
.objectid
= btrfs_ino(inode
);
170 key
.type
= BTRFS_EXTENT_DATA_KEY
;
172 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
173 path
->leave_spinning
= 1;
174 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
181 leaf
= path
->nodes
[0];
182 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
183 struct btrfs_file_extent_item
);
184 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
185 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
186 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
187 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
188 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
189 ptr
= btrfs_file_extent_inline_start(ei
);
191 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
194 while (compressed_size
> 0) {
195 cpage
= compressed_pages
[i
];
196 cur_size
= min_t(unsigned long, compressed_size
,
199 kaddr
= kmap_atomic(cpage
);
200 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
201 kunmap_atomic(kaddr
);
205 compressed_size
-= cur_size
;
207 btrfs_set_file_extent_compression(leaf
, ei
,
210 page
= find_get_page(inode
->i_mapping
,
211 start
>> PAGE_SHIFT
);
212 btrfs_set_file_extent_compression(leaf
, ei
, 0);
213 kaddr
= kmap_atomic(page
);
214 offset
= start
& (PAGE_SIZE
- 1);
215 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
216 kunmap_atomic(kaddr
);
219 btrfs_mark_buffer_dirty(leaf
);
220 btrfs_release_path(path
);
223 * we're an inline extent, so nobody can
224 * extend the file past i_size without locking
225 * a page we already have locked.
227 * We must do any isize and inode updates
228 * before we unlock the pages. Otherwise we
229 * could end up racing with unlink.
231 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
232 ret
= btrfs_update_inode(trans
, root
, inode
);
241 * conditionally insert an inline extent into the file. This
242 * does the checks required to make sure the data is small enough
243 * to fit as an inline extent.
245 static noinline
int cow_file_range_inline(struct btrfs_root
*root
,
246 struct inode
*inode
, u64 start
,
247 u64 end
, size_t compressed_size
,
249 struct page
**compressed_pages
)
251 struct btrfs_trans_handle
*trans
;
252 u64 isize
= i_size_read(inode
);
253 u64 actual_end
= min(end
+ 1, isize
);
254 u64 inline_len
= actual_end
- start
;
255 u64 aligned_end
= ALIGN(end
, root
->sectorsize
);
256 u64 data_len
= inline_len
;
258 struct btrfs_path
*path
;
259 int extent_inserted
= 0;
260 u32 extent_item_size
;
263 data_len
= compressed_size
;
266 actual_end
> root
->sectorsize
||
267 data_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
269 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
271 data_len
> root
->fs_info
->max_inline
) {
275 path
= btrfs_alloc_path();
279 trans
= btrfs_join_transaction(root
);
281 btrfs_free_path(path
);
282 return PTR_ERR(trans
);
284 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
286 if (compressed_size
&& compressed_pages
)
287 extent_item_size
= btrfs_file_extent_calc_inline_size(
290 extent_item_size
= btrfs_file_extent_calc_inline_size(
293 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
,
294 start
, aligned_end
, NULL
,
295 1, 1, extent_item_size
, &extent_inserted
);
297 btrfs_abort_transaction(trans
, root
, ret
);
301 if (isize
> actual_end
)
302 inline_len
= min_t(u64
, isize
, actual_end
);
303 ret
= insert_inline_extent(trans
, path
, extent_inserted
,
305 inline_len
, compressed_size
,
306 compress_type
, compressed_pages
);
307 if (ret
&& ret
!= -ENOSPC
) {
308 btrfs_abort_transaction(trans
, root
, ret
);
310 } else if (ret
== -ENOSPC
) {
315 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
316 btrfs_delalloc_release_metadata(inode
, end
+ 1 - start
);
317 btrfs_drop_extent_cache(inode
, start
, aligned_end
- 1, 0);
320 * Don't forget to free the reserved space, as for inlined extent
321 * it won't count as data extent, free them directly here.
322 * And at reserve time, it's always aligned to page size, so
323 * just free one page here.
325 btrfs_qgroup_free_data(inode
, 0, PAGE_SIZE
);
326 btrfs_free_path(path
);
327 btrfs_end_transaction(trans
, root
);
331 struct async_extent
{
336 unsigned long nr_pages
;
338 struct list_head list
;
343 struct btrfs_root
*root
;
344 struct page
*locked_page
;
347 struct list_head extents
;
348 struct btrfs_work work
;
351 static noinline
int add_async_extent(struct async_cow
*cow
,
352 u64 start
, u64 ram_size
,
355 unsigned long nr_pages
,
358 struct async_extent
*async_extent
;
360 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
361 BUG_ON(!async_extent
); /* -ENOMEM */
362 async_extent
->start
= start
;
363 async_extent
->ram_size
= ram_size
;
364 async_extent
->compressed_size
= compressed_size
;
365 async_extent
->pages
= pages
;
366 async_extent
->nr_pages
= nr_pages
;
367 async_extent
->compress_type
= compress_type
;
368 list_add_tail(&async_extent
->list
, &cow
->extents
);
372 static inline int inode_need_compress(struct inode
*inode
)
374 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
377 if (btrfs_test_opt(root
, FORCE_COMPRESS
))
379 /* bad compression ratios */
380 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
382 if (btrfs_test_opt(root
, COMPRESS
) ||
383 BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
||
384 BTRFS_I(inode
)->force_compress
)
390 * we create compressed extents in two phases. The first
391 * phase compresses a range of pages that have already been
392 * locked (both pages and state bits are locked).
394 * This is done inside an ordered work queue, and the compression
395 * is spread across many cpus. The actual IO submission is step
396 * two, and the ordered work queue takes care of making sure that
397 * happens in the same order things were put onto the queue by
398 * writepages and friends.
400 * If this code finds it can't get good compression, it puts an
401 * entry onto the work queue to write the uncompressed bytes. This
402 * makes sure that both compressed inodes and uncompressed inodes
403 * are written in the same order that the flusher thread sent them
406 static noinline
void compress_file_range(struct inode
*inode
,
407 struct page
*locked_page
,
409 struct async_cow
*async_cow
,
412 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
414 u64 blocksize
= root
->sectorsize
;
416 u64 isize
= i_size_read(inode
);
418 struct page
**pages
= NULL
;
419 unsigned long nr_pages
;
420 unsigned long nr_pages_ret
= 0;
421 unsigned long total_compressed
= 0;
422 unsigned long total_in
= 0;
423 unsigned long max_compressed
= SZ_128K
;
424 unsigned long max_uncompressed
= SZ_128K
;
427 int compress_type
= root
->fs_info
->compress_type
;
430 /* if this is a small write inside eof, kick off a defrag */
431 if ((end
- start
+ 1) < SZ_16K
&&
432 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
433 btrfs_add_inode_defrag(NULL
, inode
);
435 actual_end
= min_t(u64
, isize
, end
+ 1);
438 nr_pages
= (end
>> PAGE_SHIFT
) - (start
>> PAGE_SHIFT
) + 1;
439 nr_pages
= min_t(unsigned long, nr_pages
, SZ_128K
/ PAGE_SIZE
);
442 * we don't want to send crud past the end of i_size through
443 * compression, that's just a waste of CPU time. So, if the
444 * end of the file is before the start of our current
445 * requested range of bytes, we bail out to the uncompressed
446 * cleanup code that can deal with all of this.
448 * It isn't really the fastest way to fix things, but this is a
449 * very uncommon corner.
451 if (actual_end
<= start
)
452 goto cleanup_and_bail_uncompressed
;
454 total_compressed
= actual_end
- start
;
457 * skip compression for a small file range(<=blocksize) that
458 * isn't an inline extent, since it doesn't save disk space at all.
460 if (total_compressed
<= blocksize
&&
461 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
462 goto cleanup_and_bail_uncompressed
;
464 /* we want to make sure that amount of ram required to uncompress
465 * an extent is reasonable, so we limit the total size in ram
466 * of a compressed extent to 128k. This is a crucial number
467 * because it also controls how easily we can spread reads across
468 * cpus for decompression.
470 * We also want to make sure the amount of IO required to do
471 * a random read is reasonably small, so we limit the size of
472 * a compressed extent to 128k.
474 total_compressed
= min(total_compressed
, max_uncompressed
);
475 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
476 num_bytes
= max(blocksize
, num_bytes
);
481 * we do compression for mount -o compress and when the
482 * inode has not been flagged as nocompress. This flag can
483 * change at any time if we discover bad compression ratios.
485 if (inode_need_compress(inode
)) {
487 pages
= kcalloc(nr_pages
, sizeof(struct page
*), GFP_NOFS
);
489 /* just bail out to the uncompressed code */
493 if (BTRFS_I(inode
)->force_compress
)
494 compress_type
= BTRFS_I(inode
)->force_compress
;
497 * we need to call clear_page_dirty_for_io on each
498 * page in the range. Otherwise applications with the file
499 * mmap'd can wander in and change the page contents while
500 * we are compressing them.
502 * If the compression fails for any reason, we set the pages
503 * dirty again later on.
505 extent_range_clear_dirty_for_io(inode
, start
, end
);
507 ret
= btrfs_compress_pages(compress_type
,
508 inode
->i_mapping
, start
,
509 total_compressed
, pages
,
510 nr_pages
, &nr_pages_ret
,
516 unsigned long offset
= total_compressed
&
518 struct page
*page
= pages
[nr_pages_ret
- 1];
521 /* zero the tail end of the last page, we might be
522 * sending it down to disk
525 kaddr
= kmap_atomic(page
);
526 memset(kaddr
+ offset
, 0,
528 kunmap_atomic(kaddr
);
535 /* lets try to make an inline extent */
536 if (ret
|| total_in
< (actual_end
- start
)) {
537 /* we didn't compress the entire range, try
538 * to make an uncompressed inline extent.
540 ret
= cow_file_range_inline(root
, inode
, start
, end
,
543 /* try making a compressed inline extent */
544 ret
= cow_file_range_inline(root
, inode
, start
, end
,
546 compress_type
, pages
);
549 unsigned long clear_flags
= EXTENT_DELALLOC
|
551 unsigned long page_error_op
;
553 clear_flags
|= (ret
< 0) ? EXTENT_DO_ACCOUNTING
: 0;
554 page_error_op
= ret
< 0 ? PAGE_SET_ERROR
: 0;
557 * inline extent creation worked or returned error,
558 * we don't need to create any more async work items.
559 * Unlock and free up our temp pages.
561 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
,
562 clear_flags
, PAGE_UNLOCK
|
573 * we aren't doing an inline extent round the compressed size
574 * up to a block size boundary so the allocator does sane
577 total_compressed
= ALIGN(total_compressed
, blocksize
);
580 * one last check to make sure the compression is really a
581 * win, compare the page count read with the blocks on disk
583 total_in
= ALIGN(total_in
, PAGE_SIZE
);
584 if (total_compressed
>= total_in
) {
587 num_bytes
= total_in
;
590 if (!will_compress
&& pages
) {
592 * the compression code ran but failed to make things smaller,
593 * free any pages it allocated and our page pointer array
595 for (i
= 0; i
< nr_pages_ret
; i
++) {
596 WARN_ON(pages
[i
]->mapping
);
601 total_compressed
= 0;
604 /* flag the file so we don't compress in the future */
605 if (!btrfs_test_opt(root
, FORCE_COMPRESS
) &&
606 !(BTRFS_I(inode
)->force_compress
)) {
607 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
613 /* the async work queues will take care of doing actual
614 * allocation on disk for these compressed pages,
615 * and will submit them to the elevator.
617 add_async_extent(async_cow
, start
, num_bytes
,
618 total_compressed
, pages
, nr_pages_ret
,
621 if (start
+ num_bytes
< end
) {
628 cleanup_and_bail_uncompressed
:
630 * No compression, but we still need to write the pages in
631 * the file we've been given so far. redirty the locked
632 * page if it corresponds to our extent and set things up
633 * for the async work queue to run cow_file_range to do
634 * the normal delalloc dance
636 if (page_offset(locked_page
) >= start
&&
637 page_offset(locked_page
) <= end
) {
638 __set_page_dirty_nobuffers(locked_page
);
639 /* unlocked later on in the async handlers */
642 extent_range_redirty_for_io(inode
, start
, end
);
643 add_async_extent(async_cow
, start
, end
- start
+ 1,
644 0, NULL
, 0, BTRFS_COMPRESS_NONE
);
651 for (i
= 0; i
< nr_pages_ret
; i
++) {
652 WARN_ON(pages
[i
]->mapping
);
658 static void free_async_extent_pages(struct async_extent
*async_extent
)
662 if (!async_extent
->pages
)
665 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
666 WARN_ON(async_extent
->pages
[i
]->mapping
);
667 put_page(async_extent
->pages
[i
]);
669 kfree(async_extent
->pages
);
670 async_extent
->nr_pages
= 0;
671 async_extent
->pages
= NULL
;
675 * phase two of compressed writeback. This is the ordered portion
676 * of the code, which only gets called in the order the work was
677 * queued. We walk all the async extents created by compress_file_range
678 * and send them down to the disk.
680 static noinline
void submit_compressed_extents(struct inode
*inode
,
681 struct async_cow
*async_cow
)
683 struct async_extent
*async_extent
;
685 struct btrfs_key ins
;
686 struct extent_map
*em
;
687 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
688 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
689 struct extent_io_tree
*io_tree
;
693 while (!list_empty(&async_cow
->extents
)) {
694 async_extent
= list_entry(async_cow
->extents
.next
,
695 struct async_extent
, list
);
696 list_del(&async_extent
->list
);
698 io_tree
= &BTRFS_I(inode
)->io_tree
;
701 /* did the compression code fall back to uncompressed IO? */
702 if (!async_extent
->pages
) {
703 int page_started
= 0;
704 unsigned long nr_written
= 0;
706 lock_extent(io_tree
, async_extent
->start
,
707 async_extent
->start
+
708 async_extent
->ram_size
- 1);
710 /* allocate blocks */
711 ret
= cow_file_range(inode
, async_cow
->locked_page
,
713 async_extent
->start
+
714 async_extent
->ram_size
- 1,
715 &page_started
, &nr_written
, 0);
720 * if page_started, cow_file_range inserted an
721 * inline extent and took care of all the unlocking
722 * and IO for us. Otherwise, we need to submit
723 * all those pages down to the drive.
725 if (!page_started
&& !ret
)
726 extent_write_locked_range(io_tree
,
727 inode
, async_extent
->start
,
728 async_extent
->start
+
729 async_extent
->ram_size
- 1,
733 unlock_page(async_cow
->locked_page
);
739 lock_extent(io_tree
, async_extent
->start
,
740 async_extent
->start
+ async_extent
->ram_size
- 1);
742 ret
= btrfs_reserve_extent(root
,
743 async_extent
->compressed_size
,
744 async_extent
->compressed_size
,
745 0, alloc_hint
, &ins
, 1, 1);
747 free_async_extent_pages(async_extent
);
749 if (ret
== -ENOSPC
) {
750 unlock_extent(io_tree
, async_extent
->start
,
751 async_extent
->start
+
752 async_extent
->ram_size
- 1);
755 * we need to redirty the pages if we decide to
756 * fallback to uncompressed IO, otherwise we
757 * will not submit these pages down to lower
760 extent_range_redirty_for_io(inode
,
762 async_extent
->start
+
763 async_extent
->ram_size
- 1);
770 * here we're doing allocation and writeback of the
773 btrfs_drop_extent_cache(inode
, async_extent
->start
,
774 async_extent
->start
+
775 async_extent
->ram_size
- 1, 0);
777 em
= alloc_extent_map();
780 goto out_free_reserve
;
782 em
->start
= async_extent
->start
;
783 em
->len
= async_extent
->ram_size
;
784 em
->orig_start
= em
->start
;
785 em
->mod_start
= em
->start
;
786 em
->mod_len
= em
->len
;
788 em
->block_start
= ins
.objectid
;
789 em
->block_len
= ins
.offset
;
790 em
->orig_block_len
= ins
.offset
;
791 em
->ram_bytes
= async_extent
->ram_size
;
792 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
793 em
->compress_type
= async_extent
->compress_type
;
794 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
795 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
799 write_lock(&em_tree
->lock
);
800 ret
= add_extent_mapping(em_tree
, em
, 1);
801 write_unlock(&em_tree
->lock
);
802 if (ret
!= -EEXIST
) {
806 btrfs_drop_extent_cache(inode
, async_extent
->start
,
807 async_extent
->start
+
808 async_extent
->ram_size
- 1, 0);
812 goto out_free_reserve
;
814 ret
= btrfs_add_ordered_extent_compress(inode
,
817 async_extent
->ram_size
,
819 BTRFS_ORDERED_COMPRESSED
,
820 async_extent
->compress_type
);
822 btrfs_drop_extent_cache(inode
, async_extent
->start
,
823 async_extent
->start
+
824 async_extent
->ram_size
- 1, 0);
825 goto out_free_reserve
;
827 btrfs_dec_block_group_reservations(root
->fs_info
, ins
.objectid
);
830 * clear dirty, set writeback and unlock the pages.
832 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
833 async_extent
->start
+
834 async_extent
->ram_size
- 1,
835 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
,
836 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
838 ret
= btrfs_submit_compressed_write(inode
,
840 async_extent
->ram_size
,
842 ins
.offset
, async_extent
->pages
,
843 async_extent
->nr_pages
);
845 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
846 struct page
*p
= async_extent
->pages
[0];
847 const u64 start
= async_extent
->start
;
848 const u64 end
= start
+ async_extent
->ram_size
- 1;
850 p
->mapping
= inode
->i_mapping
;
851 tree
->ops
->writepage_end_io_hook(p
, start
, end
,
854 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
, 0,
857 free_async_extent_pages(async_extent
);
859 alloc_hint
= ins
.objectid
+ ins
.offset
;
865 btrfs_dec_block_group_reservations(root
->fs_info
, ins
.objectid
);
866 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
868 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
869 async_extent
->start
+
870 async_extent
->ram_size
- 1,
871 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
|
872 EXTENT_DEFRAG
| EXTENT_DO_ACCOUNTING
,
873 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
874 PAGE_SET_WRITEBACK
| PAGE_END_WRITEBACK
|
876 free_async_extent_pages(async_extent
);
881 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
884 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
885 struct extent_map
*em
;
888 read_lock(&em_tree
->lock
);
889 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
892 * if block start isn't an actual block number then find the
893 * first block in this inode and use that as a hint. If that
894 * block is also bogus then just don't worry about it.
896 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
898 em
= search_extent_mapping(em_tree
, 0, 0);
899 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
900 alloc_hint
= em
->block_start
;
904 alloc_hint
= em
->block_start
;
908 read_unlock(&em_tree
->lock
);
914 * when extent_io.c finds a delayed allocation range in the file,
915 * the call backs end up in this code. The basic idea is to
916 * allocate extents on disk for the range, and create ordered data structs
917 * in ram to track those extents.
919 * locked_page is the page that writepage had locked already. We use
920 * it to make sure we don't do extra locks or unlocks.
922 * *page_started is set to one if we unlock locked_page and do everything
923 * required to start IO on it. It may be clean and already done with
926 static noinline
int cow_file_range(struct inode
*inode
,
927 struct page
*locked_page
,
928 u64 start
, u64 end
, int *page_started
,
929 unsigned long *nr_written
,
932 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
935 unsigned long ram_size
;
938 u64 blocksize
= root
->sectorsize
;
939 struct btrfs_key ins
;
940 struct extent_map
*em
;
941 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
944 if (btrfs_is_free_space_inode(inode
)) {
950 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
951 num_bytes
= max(blocksize
, num_bytes
);
952 disk_num_bytes
= num_bytes
;
954 /* if this is a small write inside eof, kick off defrag */
955 if (num_bytes
< SZ_64K
&&
956 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
957 btrfs_add_inode_defrag(NULL
, inode
);
960 /* lets try to make an inline extent */
961 ret
= cow_file_range_inline(root
, inode
, start
, end
, 0, 0,
964 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
,
965 EXTENT_LOCKED
| EXTENT_DELALLOC
|
966 EXTENT_DEFRAG
, PAGE_UNLOCK
|
967 PAGE_CLEAR_DIRTY
| PAGE_SET_WRITEBACK
|
970 *nr_written
= *nr_written
+
971 (end
- start
+ PAGE_SIZE
) / PAGE_SIZE
;
974 } else if (ret
< 0) {
979 BUG_ON(disk_num_bytes
>
980 btrfs_super_total_bytes(root
->fs_info
->super_copy
));
982 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
983 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
985 while (disk_num_bytes
> 0) {
988 cur_alloc_size
= disk_num_bytes
;
989 ret
= btrfs_reserve_extent(root
, cur_alloc_size
,
990 root
->sectorsize
, 0, alloc_hint
,
995 em
= alloc_extent_map();
1001 em
->orig_start
= em
->start
;
1002 ram_size
= ins
.offset
;
1003 em
->len
= ins
.offset
;
1004 em
->mod_start
= em
->start
;
1005 em
->mod_len
= em
->len
;
1007 em
->block_start
= ins
.objectid
;
1008 em
->block_len
= ins
.offset
;
1009 em
->orig_block_len
= ins
.offset
;
1010 em
->ram_bytes
= ram_size
;
1011 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1012 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1013 em
->generation
= -1;
1016 write_lock(&em_tree
->lock
);
1017 ret
= add_extent_mapping(em_tree
, em
, 1);
1018 write_unlock(&em_tree
->lock
);
1019 if (ret
!= -EEXIST
) {
1020 free_extent_map(em
);
1023 btrfs_drop_extent_cache(inode
, start
,
1024 start
+ ram_size
- 1, 0);
1029 cur_alloc_size
= ins
.offset
;
1030 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
1031 ram_size
, cur_alloc_size
, 0);
1033 goto out_drop_extent_cache
;
1035 if (root
->root_key
.objectid
==
1036 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1037 ret
= btrfs_reloc_clone_csums(inode
, start
,
1040 goto out_drop_extent_cache
;
1043 btrfs_dec_block_group_reservations(root
->fs_info
, ins
.objectid
);
1045 if (disk_num_bytes
< cur_alloc_size
)
1048 /* we're not doing compressed IO, don't unlock the first
1049 * page (which the caller expects to stay locked), don't
1050 * clear any dirty bits and don't set any writeback bits
1052 * Do set the Private2 bit so we know this page was properly
1053 * setup for writepage
1055 op
= unlock
? PAGE_UNLOCK
: 0;
1056 op
|= PAGE_SET_PRIVATE2
;
1058 extent_clear_unlock_delalloc(inode
, start
,
1059 start
+ ram_size
- 1, locked_page
,
1060 EXTENT_LOCKED
| EXTENT_DELALLOC
,
1062 disk_num_bytes
-= cur_alloc_size
;
1063 num_bytes
-= cur_alloc_size
;
1064 alloc_hint
= ins
.objectid
+ ins
.offset
;
1065 start
+= cur_alloc_size
;
1070 out_drop_extent_cache
:
1071 btrfs_drop_extent_cache(inode
, start
, start
+ ram_size
- 1, 0);
1073 btrfs_dec_block_group_reservations(root
->fs_info
, ins
.objectid
);
1074 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
1076 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1077 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
1078 EXTENT_DELALLOC
| EXTENT_DEFRAG
,
1079 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
1080 PAGE_SET_WRITEBACK
| PAGE_END_WRITEBACK
);
1085 * work queue call back to started compression on a file and pages
1087 static noinline
void async_cow_start(struct btrfs_work
*work
)
1089 struct async_cow
*async_cow
;
1091 async_cow
= container_of(work
, struct async_cow
, work
);
1093 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
1094 async_cow
->start
, async_cow
->end
, async_cow
,
1096 if (num_added
== 0) {
1097 btrfs_add_delayed_iput(async_cow
->inode
);
1098 async_cow
->inode
= NULL
;
1103 * work queue call back to submit previously compressed pages
1105 static noinline
void async_cow_submit(struct btrfs_work
*work
)
1107 struct async_cow
*async_cow
;
1108 struct btrfs_root
*root
;
1109 unsigned long nr_pages
;
1111 async_cow
= container_of(work
, struct async_cow
, work
);
1113 root
= async_cow
->root
;
1114 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_SIZE
) >>
1118 * atomic_sub_return implies a barrier for waitqueue_active
1120 if (atomic_sub_return(nr_pages
, &root
->fs_info
->async_delalloc_pages
) <
1122 waitqueue_active(&root
->fs_info
->async_submit_wait
))
1123 wake_up(&root
->fs_info
->async_submit_wait
);
1125 if (async_cow
->inode
)
1126 submit_compressed_extents(async_cow
->inode
, async_cow
);
1129 static noinline
void async_cow_free(struct btrfs_work
*work
)
1131 struct async_cow
*async_cow
;
1132 async_cow
= container_of(work
, struct async_cow
, work
);
1133 if (async_cow
->inode
)
1134 btrfs_add_delayed_iput(async_cow
->inode
);
1138 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
1139 u64 start
, u64 end
, int *page_started
,
1140 unsigned long *nr_written
)
1142 struct async_cow
*async_cow
;
1143 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1144 unsigned long nr_pages
;
1146 int limit
= 10 * SZ_1M
;
1148 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
1149 1, 0, NULL
, GFP_NOFS
);
1150 while (start
< end
) {
1151 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
1152 BUG_ON(!async_cow
); /* -ENOMEM */
1153 async_cow
->inode
= igrab(inode
);
1154 async_cow
->root
= root
;
1155 async_cow
->locked_page
= locked_page
;
1156 async_cow
->start
= start
;
1158 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
&&
1159 !btrfs_test_opt(root
, FORCE_COMPRESS
))
1162 cur_end
= min(end
, start
+ SZ_512K
- 1);
1164 async_cow
->end
= cur_end
;
1165 INIT_LIST_HEAD(&async_cow
->extents
);
1167 btrfs_init_work(&async_cow
->work
,
1168 btrfs_delalloc_helper
,
1169 async_cow_start
, async_cow_submit
,
1172 nr_pages
= (cur_end
- start
+ PAGE_SIZE
) >>
1174 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
1176 btrfs_queue_work(root
->fs_info
->delalloc_workers
,
1179 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
1180 wait_event(root
->fs_info
->async_submit_wait
,
1181 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
1185 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
1186 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
1187 wait_event(root
->fs_info
->async_submit_wait
,
1188 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
1192 *nr_written
+= nr_pages
;
1193 start
= cur_end
+ 1;
1199 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1200 u64 bytenr
, u64 num_bytes
)
1203 struct btrfs_ordered_sum
*sums
;
1206 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1207 bytenr
+ num_bytes
- 1, &list
, 0);
1208 if (ret
== 0 && list_empty(&list
))
1211 while (!list_empty(&list
)) {
1212 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1213 list_del(&sums
->list
);
1220 * when nowcow writeback call back. This checks for snapshots or COW copies
1221 * of the extents that exist in the file, and COWs the file as required.
1223 * If no cow copies or snapshots exist, we write directly to the existing
1226 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1227 struct page
*locked_page
,
1228 u64 start
, u64 end
, int *page_started
, int force
,
1229 unsigned long *nr_written
)
1231 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1232 struct btrfs_trans_handle
*trans
;
1233 struct extent_buffer
*leaf
;
1234 struct btrfs_path
*path
;
1235 struct btrfs_file_extent_item
*fi
;
1236 struct btrfs_key found_key
;
1251 u64 ino
= btrfs_ino(inode
);
1253 path
= btrfs_alloc_path();
1255 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1256 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1257 EXTENT_DO_ACCOUNTING
|
1258 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1260 PAGE_SET_WRITEBACK
|
1261 PAGE_END_WRITEBACK
);
1265 nolock
= btrfs_is_free_space_inode(inode
);
1268 trans
= btrfs_join_transaction_nolock(root
);
1270 trans
= btrfs_join_transaction(root
);
1272 if (IS_ERR(trans
)) {
1273 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1274 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1275 EXTENT_DO_ACCOUNTING
|
1276 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1278 PAGE_SET_WRITEBACK
|
1279 PAGE_END_WRITEBACK
);
1280 btrfs_free_path(path
);
1281 return PTR_ERR(trans
);
1284 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1286 cow_start
= (u64
)-1;
1289 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
1293 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1294 leaf
= path
->nodes
[0];
1295 btrfs_item_key_to_cpu(leaf
, &found_key
,
1296 path
->slots
[0] - 1);
1297 if (found_key
.objectid
== ino
&&
1298 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1303 leaf
= path
->nodes
[0];
1304 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1305 ret
= btrfs_next_leaf(root
, path
);
1310 leaf
= path
->nodes
[0];
1316 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1318 if (found_key
.objectid
> ino
)
1320 if (WARN_ON_ONCE(found_key
.objectid
< ino
) ||
1321 found_key
.type
< BTRFS_EXTENT_DATA_KEY
) {
1325 if (found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1326 found_key
.offset
> end
)
1329 if (found_key
.offset
> cur_offset
) {
1330 extent_end
= found_key
.offset
;
1335 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1336 struct btrfs_file_extent_item
);
1337 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1339 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
1340 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1341 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1342 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1343 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1344 extent_end
= found_key
.offset
+
1345 btrfs_file_extent_num_bytes(leaf
, fi
);
1347 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1348 if (extent_end
<= start
) {
1352 if (disk_bytenr
== 0)
1354 if (btrfs_file_extent_compression(leaf
, fi
) ||
1355 btrfs_file_extent_encryption(leaf
, fi
) ||
1356 btrfs_file_extent_other_encoding(leaf
, fi
))
1358 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1360 if (btrfs_extent_readonly(root
, disk_bytenr
))
1362 if (btrfs_cross_ref_exist(trans
, root
, ino
,
1364 extent_offset
, disk_bytenr
))
1366 disk_bytenr
+= extent_offset
;
1367 disk_bytenr
+= cur_offset
- found_key
.offset
;
1368 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1370 * if there are pending snapshots for this root,
1371 * we fall into common COW way.
1374 err
= btrfs_start_write_no_snapshoting(root
);
1379 * force cow if csum exists in the range.
1380 * this ensure that csum for a given extent are
1381 * either valid or do not exist.
1383 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1385 if (!btrfs_inc_nocow_writers(root
->fs_info
,
1389 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1390 extent_end
= found_key
.offset
+
1391 btrfs_file_extent_inline_len(leaf
,
1392 path
->slots
[0], fi
);
1393 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1398 if (extent_end
<= start
) {
1400 if (!nolock
&& nocow
)
1401 btrfs_end_write_no_snapshoting(root
);
1403 btrfs_dec_nocow_writers(root
->fs_info
,
1408 if (cow_start
== (u64
)-1)
1409 cow_start
= cur_offset
;
1410 cur_offset
= extent_end
;
1411 if (cur_offset
> end
)
1417 btrfs_release_path(path
);
1418 if (cow_start
!= (u64
)-1) {
1419 ret
= cow_file_range(inode
, locked_page
,
1420 cow_start
, found_key
.offset
- 1,
1421 page_started
, nr_written
, 1);
1423 if (!nolock
&& nocow
)
1424 btrfs_end_write_no_snapshoting(root
);
1426 btrfs_dec_nocow_writers(root
->fs_info
,
1430 cow_start
= (u64
)-1;
1433 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1434 struct extent_map
*em
;
1435 struct extent_map_tree
*em_tree
;
1436 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1437 em
= alloc_extent_map();
1438 BUG_ON(!em
); /* -ENOMEM */
1439 em
->start
= cur_offset
;
1440 em
->orig_start
= found_key
.offset
- extent_offset
;
1441 em
->len
= num_bytes
;
1442 em
->block_len
= num_bytes
;
1443 em
->block_start
= disk_bytenr
;
1444 em
->orig_block_len
= disk_num_bytes
;
1445 em
->ram_bytes
= ram_bytes
;
1446 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1447 em
->mod_start
= em
->start
;
1448 em
->mod_len
= em
->len
;
1449 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1450 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
1451 em
->generation
= -1;
1453 write_lock(&em_tree
->lock
);
1454 ret
= add_extent_mapping(em_tree
, em
, 1);
1455 write_unlock(&em_tree
->lock
);
1456 if (ret
!= -EEXIST
) {
1457 free_extent_map(em
);
1460 btrfs_drop_extent_cache(inode
, em
->start
,
1461 em
->start
+ em
->len
- 1, 0);
1463 type
= BTRFS_ORDERED_PREALLOC
;
1465 type
= BTRFS_ORDERED_NOCOW
;
1468 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1469 num_bytes
, num_bytes
, type
);
1471 btrfs_dec_nocow_writers(root
->fs_info
, disk_bytenr
);
1472 BUG_ON(ret
); /* -ENOMEM */
1474 if (root
->root_key
.objectid
==
1475 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1476 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1479 if (!nolock
&& nocow
)
1480 btrfs_end_write_no_snapshoting(root
);
1485 extent_clear_unlock_delalloc(inode
, cur_offset
,
1486 cur_offset
+ num_bytes
- 1,
1487 locked_page
, EXTENT_LOCKED
|
1488 EXTENT_DELALLOC
, PAGE_UNLOCK
|
1490 if (!nolock
&& nocow
)
1491 btrfs_end_write_no_snapshoting(root
);
1492 cur_offset
= extent_end
;
1493 if (cur_offset
> end
)
1496 btrfs_release_path(path
);
1498 if (cur_offset
<= end
&& cow_start
== (u64
)-1) {
1499 cow_start
= cur_offset
;
1503 if (cow_start
!= (u64
)-1) {
1504 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1505 page_started
, nr_written
, 1);
1511 err
= btrfs_end_transaction(trans
, root
);
1515 if (ret
&& cur_offset
< end
)
1516 extent_clear_unlock_delalloc(inode
, cur_offset
, end
,
1517 locked_page
, EXTENT_LOCKED
|
1518 EXTENT_DELALLOC
| EXTENT_DEFRAG
|
1519 EXTENT_DO_ACCOUNTING
, PAGE_UNLOCK
|
1521 PAGE_SET_WRITEBACK
|
1522 PAGE_END_WRITEBACK
);
1523 btrfs_free_path(path
);
1527 static inline int need_force_cow(struct inode
*inode
, u64 start
, u64 end
)
1530 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
1531 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
))
1535 * @defrag_bytes is a hint value, no spinlock held here,
1536 * if is not zero, it means the file is defragging.
1537 * Force cow if given extent needs to be defragged.
1539 if (BTRFS_I(inode
)->defrag_bytes
&&
1540 test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1541 EXTENT_DEFRAG
, 0, NULL
))
1548 * extent_io.c call back to do delayed allocation processing
1550 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1551 u64 start
, u64 end
, int *page_started
,
1552 unsigned long *nr_written
)
1555 int force_cow
= need_force_cow(inode
, start
, end
);
1557 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
&& !force_cow
) {
1558 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1559 page_started
, 1, nr_written
);
1560 } else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
&& !force_cow
) {
1561 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1562 page_started
, 0, nr_written
);
1563 } else if (!inode_need_compress(inode
)) {
1564 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1565 page_started
, nr_written
, 1);
1567 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1568 &BTRFS_I(inode
)->runtime_flags
);
1569 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1570 page_started
, nr_written
);
1575 static void btrfs_split_extent_hook(struct inode
*inode
,
1576 struct extent_state
*orig
, u64 split
)
1580 /* not delalloc, ignore it */
1581 if (!(orig
->state
& EXTENT_DELALLOC
))
1584 size
= orig
->end
- orig
->start
+ 1;
1585 if (size
> BTRFS_MAX_EXTENT_SIZE
) {
1590 * See the explanation in btrfs_merge_extent_hook, the same
1591 * applies here, just in reverse.
1593 new_size
= orig
->end
- split
+ 1;
1594 num_extents
= div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1595 BTRFS_MAX_EXTENT_SIZE
);
1596 new_size
= split
- orig
->start
;
1597 num_extents
+= div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1598 BTRFS_MAX_EXTENT_SIZE
);
1599 if (div64_u64(size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1600 BTRFS_MAX_EXTENT_SIZE
) >= num_extents
)
1604 spin_lock(&BTRFS_I(inode
)->lock
);
1605 BTRFS_I(inode
)->outstanding_extents
++;
1606 spin_unlock(&BTRFS_I(inode
)->lock
);
1610 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1611 * extents so we can keep track of new extents that are just merged onto old
1612 * extents, such as when we are doing sequential writes, so we can properly
1613 * account for the metadata space we'll need.
1615 static void btrfs_merge_extent_hook(struct inode
*inode
,
1616 struct extent_state
*new,
1617 struct extent_state
*other
)
1619 u64 new_size
, old_size
;
1622 /* not delalloc, ignore it */
1623 if (!(other
->state
& EXTENT_DELALLOC
))
1626 if (new->start
> other
->start
)
1627 new_size
= new->end
- other
->start
+ 1;
1629 new_size
= other
->end
- new->start
+ 1;
1631 /* we're not bigger than the max, unreserve the space and go */
1632 if (new_size
<= BTRFS_MAX_EXTENT_SIZE
) {
1633 spin_lock(&BTRFS_I(inode
)->lock
);
1634 BTRFS_I(inode
)->outstanding_extents
--;
1635 spin_unlock(&BTRFS_I(inode
)->lock
);
1640 * We have to add up either side to figure out how many extents were
1641 * accounted for before we merged into one big extent. If the number of
1642 * extents we accounted for is <= the amount we need for the new range
1643 * then we can return, otherwise drop. Think of it like this
1647 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1648 * need 2 outstanding extents, on one side we have 1 and the other side
1649 * we have 1 so they are == and we can return. But in this case
1651 * [MAX_SIZE+4k][MAX_SIZE+4k]
1653 * Each range on their own accounts for 2 extents, but merged together
1654 * they are only 3 extents worth of accounting, so we need to drop in
1657 old_size
= other
->end
- other
->start
+ 1;
1658 num_extents
= div64_u64(old_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1659 BTRFS_MAX_EXTENT_SIZE
);
1660 old_size
= new->end
- new->start
+ 1;
1661 num_extents
+= div64_u64(old_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1662 BTRFS_MAX_EXTENT_SIZE
);
1664 if (div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1665 BTRFS_MAX_EXTENT_SIZE
) >= num_extents
)
1668 spin_lock(&BTRFS_I(inode
)->lock
);
1669 BTRFS_I(inode
)->outstanding_extents
--;
1670 spin_unlock(&BTRFS_I(inode
)->lock
);
1673 static void btrfs_add_delalloc_inodes(struct btrfs_root
*root
,
1674 struct inode
*inode
)
1676 spin_lock(&root
->delalloc_lock
);
1677 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1678 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1679 &root
->delalloc_inodes
);
1680 set_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1681 &BTRFS_I(inode
)->runtime_flags
);
1682 root
->nr_delalloc_inodes
++;
1683 if (root
->nr_delalloc_inodes
== 1) {
1684 spin_lock(&root
->fs_info
->delalloc_root_lock
);
1685 BUG_ON(!list_empty(&root
->delalloc_root
));
1686 list_add_tail(&root
->delalloc_root
,
1687 &root
->fs_info
->delalloc_roots
);
1688 spin_unlock(&root
->fs_info
->delalloc_root_lock
);
1691 spin_unlock(&root
->delalloc_lock
);
1694 static void btrfs_del_delalloc_inode(struct btrfs_root
*root
,
1695 struct inode
*inode
)
1697 spin_lock(&root
->delalloc_lock
);
1698 if (!list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1699 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1700 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1701 &BTRFS_I(inode
)->runtime_flags
);
1702 root
->nr_delalloc_inodes
--;
1703 if (!root
->nr_delalloc_inodes
) {
1704 spin_lock(&root
->fs_info
->delalloc_root_lock
);
1705 BUG_ON(list_empty(&root
->delalloc_root
));
1706 list_del_init(&root
->delalloc_root
);
1707 spin_unlock(&root
->fs_info
->delalloc_root_lock
);
1710 spin_unlock(&root
->delalloc_lock
);
1714 * extent_io.c set_bit_hook, used to track delayed allocation
1715 * bytes in this file, and to maintain the list of inodes that
1716 * have pending delalloc work to be done.
1718 static void btrfs_set_bit_hook(struct inode
*inode
,
1719 struct extent_state
*state
, unsigned *bits
)
1722 if ((*bits
& EXTENT_DEFRAG
) && !(*bits
& EXTENT_DELALLOC
))
1725 * set_bit and clear bit hooks normally require _irqsave/restore
1726 * but in this case, we are only testing for the DELALLOC
1727 * bit, which is only set or cleared with irqs on
1729 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1730 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1731 u64 len
= state
->end
+ 1 - state
->start
;
1732 bool do_list
= !btrfs_is_free_space_inode(inode
);
1734 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1735 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1737 spin_lock(&BTRFS_I(inode
)->lock
);
1738 BTRFS_I(inode
)->outstanding_extents
++;
1739 spin_unlock(&BTRFS_I(inode
)->lock
);
1742 /* For sanity tests */
1743 if (btrfs_test_is_dummy_root(root
))
1746 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, len
,
1747 root
->fs_info
->delalloc_batch
);
1748 spin_lock(&BTRFS_I(inode
)->lock
);
1749 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1750 if (*bits
& EXTENT_DEFRAG
)
1751 BTRFS_I(inode
)->defrag_bytes
+= len
;
1752 if (do_list
&& !test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1753 &BTRFS_I(inode
)->runtime_flags
))
1754 btrfs_add_delalloc_inodes(root
, inode
);
1755 spin_unlock(&BTRFS_I(inode
)->lock
);
1760 * extent_io.c clear_bit_hook, see set_bit_hook for why
1762 static void btrfs_clear_bit_hook(struct inode
*inode
,
1763 struct extent_state
*state
,
1766 u64 len
= state
->end
+ 1 - state
->start
;
1767 u64 num_extents
= div64_u64(len
+ BTRFS_MAX_EXTENT_SIZE
-1,
1768 BTRFS_MAX_EXTENT_SIZE
);
1770 spin_lock(&BTRFS_I(inode
)->lock
);
1771 if ((state
->state
& EXTENT_DEFRAG
) && (*bits
& EXTENT_DEFRAG
))
1772 BTRFS_I(inode
)->defrag_bytes
-= len
;
1773 spin_unlock(&BTRFS_I(inode
)->lock
);
1776 * set_bit and clear bit hooks normally require _irqsave/restore
1777 * but in this case, we are only testing for the DELALLOC
1778 * bit, which is only set or cleared with irqs on
1780 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1781 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1782 bool do_list
= !btrfs_is_free_space_inode(inode
);
1784 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1785 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1786 } else if (!(*bits
& EXTENT_DO_ACCOUNTING
)) {
1787 spin_lock(&BTRFS_I(inode
)->lock
);
1788 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
1789 spin_unlock(&BTRFS_I(inode
)->lock
);
1793 * We don't reserve metadata space for space cache inodes so we
1794 * don't need to call dellalloc_release_metadata if there is an
1797 if (*bits
& EXTENT_DO_ACCOUNTING
&&
1798 root
!= root
->fs_info
->tree_root
)
1799 btrfs_delalloc_release_metadata(inode
, len
);
1801 /* For sanity tests. */
1802 if (btrfs_test_is_dummy_root(root
))
1805 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1806 && do_list
&& !(state
->state
& EXTENT_NORESERVE
))
1807 btrfs_free_reserved_data_space_noquota(inode
,
1810 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, -len
,
1811 root
->fs_info
->delalloc_batch
);
1812 spin_lock(&BTRFS_I(inode
)->lock
);
1813 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1814 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1815 test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1816 &BTRFS_I(inode
)->runtime_flags
))
1817 btrfs_del_delalloc_inode(root
, inode
);
1818 spin_unlock(&BTRFS_I(inode
)->lock
);
1823 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1824 * we don't create bios that span stripes or chunks
1826 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1827 size_t size
, struct bio
*bio
,
1828 unsigned long bio_flags
)
1830 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1831 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
1836 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1839 length
= bio
->bi_iter
.bi_size
;
1840 map_length
= length
;
1841 ret
= btrfs_map_block(root
->fs_info
, bio_op(bio
), logical
,
1842 &map_length
, NULL
, 0);
1843 /* Will always return 0 with map_multi == NULL */
1845 if (map_length
< length
+ size
)
1851 * in order to insert checksums into the metadata in large chunks,
1852 * we wait until bio submission time. All the pages in the bio are
1853 * checksummed and sums are attached onto the ordered extent record.
1855 * At IO completion time the cums attached on the ordered extent record
1856 * are inserted into the btree
1858 static int __btrfs_submit_bio_start(struct inode
*inode
, struct bio
*bio
,
1859 int mirror_num
, unsigned long bio_flags
,
1862 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1865 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1866 BUG_ON(ret
); /* -ENOMEM */
1871 * in order to insert checksums into the metadata in large chunks,
1872 * we wait until bio submission time. All the pages in the bio are
1873 * checksummed and sums are attached onto the ordered extent record.
1875 * At IO completion time the cums attached on the ordered extent record
1876 * are inserted into the btree
1878 static int __btrfs_submit_bio_done(struct inode
*inode
, struct bio
*bio
,
1879 int mirror_num
, unsigned long bio_flags
,
1882 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1885 ret
= btrfs_map_bio(root
, bio
, mirror_num
, 1);
1887 bio
->bi_error
= ret
;
1894 * extent_io.c submission hook. This does the right thing for csum calculation
1895 * on write, or reading the csums from the tree before a read
1897 static int btrfs_submit_bio_hook(struct inode
*inode
, struct bio
*bio
,
1898 int mirror_num
, unsigned long bio_flags
,
1901 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1902 enum btrfs_wq_endio_type metadata
= BTRFS_WQ_ENDIO_DATA
;
1905 int async
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
1907 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1909 if (btrfs_is_free_space_inode(inode
))
1910 metadata
= BTRFS_WQ_ENDIO_FREE_SPACE
;
1912 if (bio_op(bio
) != REQ_OP_WRITE
) {
1913 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, metadata
);
1917 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1918 ret
= btrfs_submit_compressed_read(inode
, bio
,
1922 } else if (!skip_sum
) {
1923 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1928 } else if (async
&& !skip_sum
) {
1929 /* csum items have already been cloned */
1930 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1932 /* we're doing a write, do the async checksumming */
1933 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1934 inode
, bio
, mirror_num
,
1935 bio_flags
, bio_offset
,
1936 __btrfs_submit_bio_start
,
1937 __btrfs_submit_bio_done
);
1939 } else if (!skip_sum
) {
1940 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1946 ret
= btrfs_map_bio(root
, bio
, mirror_num
, 0);
1950 bio
->bi_error
= ret
;
1957 * given a list of ordered sums record them in the inode. This happens
1958 * at IO completion time based on sums calculated at bio submission time.
1960 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1961 struct inode
*inode
, u64 file_offset
,
1962 struct list_head
*list
)
1964 struct btrfs_ordered_sum
*sum
;
1966 list_for_each_entry(sum
, list
, list
) {
1967 trans
->adding_csums
= 1;
1968 btrfs_csum_file_blocks(trans
,
1969 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1970 trans
->adding_csums
= 0;
1975 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1976 struct extent_state
**cached_state
)
1978 WARN_ON((end
& (PAGE_SIZE
- 1)) == 0);
1979 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1983 /* see btrfs_writepage_start_hook for details on why this is required */
1984 struct btrfs_writepage_fixup
{
1986 struct btrfs_work work
;
1989 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1991 struct btrfs_writepage_fixup
*fixup
;
1992 struct btrfs_ordered_extent
*ordered
;
1993 struct extent_state
*cached_state
= NULL
;
1995 struct inode
*inode
;
2000 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
2004 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
2005 ClearPageChecked(page
);
2009 inode
= page
->mapping
->host
;
2010 page_start
= page_offset(page
);
2011 page_end
= page_offset(page
) + PAGE_SIZE
- 1;
2013 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2016 /* already ordered? We're done */
2017 if (PagePrivate2(page
))
2020 ordered
= btrfs_lookup_ordered_range(inode
, page_start
,
2023 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
2024 page_end
, &cached_state
, GFP_NOFS
);
2026 btrfs_start_ordered_extent(inode
, ordered
, 1);
2027 btrfs_put_ordered_extent(ordered
);
2031 ret
= btrfs_delalloc_reserve_space(inode
, page_start
,
2034 mapping_set_error(page
->mapping
, ret
);
2035 end_extent_writepage(page
, ret
, page_start
, page_end
);
2036 ClearPageChecked(page
);
2040 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
2041 ClearPageChecked(page
);
2042 set_page_dirty(page
);
2044 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2045 &cached_state
, GFP_NOFS
);
2053 * There are a few paths in the higher layers of the kernel that directly
2054 * set the page dirty bit without asking the filesystem if it is a
2055 * good idea. This causes problems because we want to make sure COW
2056 * properly happens and the data=ordered rules are followed.
2058 * In our case any range that doesn't have the ORDERED bit set
2059 * hasn't been properly setup for IO. We kick off an async process
2060 * to fix it up. The async helper will wait for ordered extents, set
2061 * the delalloc bit and make it safe to write the page.
2063 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
2065 struct inode
*inode
= page
->mapping
->host
;
2066 struct btrfs_writepage_fixup
*fixup
;
2067 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2069 /* this page is properly in the ordered list */
2070 if (TestClearPagePrivate2(page
))
2073 if (PageChecked(page
))
2076 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
2080 SetPageChecked(page
);
2082 btrfs_init_work(&fixup
->work
, btrfs_fixup_helper
,
2083 btrfs_writepage_fixup_worker
, NULL
, NULL
);
2085 btrfs_queue_work(root
->fs_info
->fixup_workers
, &fixup
->work
);
2089 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
2090 struct inode
*inode
, u64 file_pos
,
2091 u64 disk_bytenr
, u64 disk_num_bytes
,
2092 u64 num_bytes
, u64 ram_bytes
,
2093 u8 compression
, u8 encryption
,
2094 u16 other_encoding
, int extent_type
)
2096 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2097 struct btrfs_file_extent_item
*fi
;
2098 struct btrfs_path
*path
;
2099 struct extent_buffer
*leaf
;
2100 struct btrfs_key ins
;
2101 int extent_inserted
= 0;
2104 path
= btrfs_alloc_path();
2109 * we may be replacing one extent in the tree with another.
2110 * The new extent is pinned in the extent map, and we don't want
2111 * to drop it from the cache until it is completely in the btree.
2113 * So, tell btrfs_drop_extents to leave this extent in the cache.
2114 * the caller is expected to unpin it and allow it to be merged
2117 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
, file_pos
,
2118 file_pos
+ num_bytes
, NULL
, 0,
2119 1, sizeof(*fi
), &extent_inserted
);
2123 if (!extent_inserted
) {
2124 ins
.objectid
= btrfs_ino(inode
);
2125 ins
.offset
= file_pos
;
2126 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
2128 path
->leave_spinning
= 1;
2129 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
2134 leaf
= path
->nodes
[0];
2135 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2136 struct btrfs_file_extent_item
);
2137 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
2138 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
2139 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
2140 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
2141 btrfs_set_file_extent_offset(leaf
, fi
, 0);
2142 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
2143 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
2144 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
2145 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
2146 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
2148 btrfs_mark_buffer_dirty(leaf
);
2149 btrfs_release_path(path
);
2151 inode_add_bytes(inode
, num_bytes
);
2153 ins
.objectid
= disk_bytenr
;
2154 ins
.offset
= disk_num_bytes
;
2155 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2156 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
2157 root
->root_key
.objectid
,
2158 btrfs_ino(inode
), file_pos
,
2161 * Release the reserved range from inode dirty range map, as it is
2162 * already moved into delayed_ref_head
2164 btrfs_qgroup_release_data(inode
, file_pos
, ram_bytes
);
2166 btrfs_free_path(path
);
2171 /* snapshot-aware defrag */
2172 struct sa_defrag_extent_backref
{
2173 struct rb_node node
;
2174 struct old_sa_defrag_extent
*old
;
2183 struct old_sa_defrag_extent
{
2184 struct list_head list
;
2185 struct new_sa_defrag_extent
*new;
2194 struct new_sa_defrag_extent
{
2195 struct rb_root root
;
2196 struct list_head head
;
2197 struct btrfs_path
*path
;
2198 struct inode
*inode
;
2206 static int backref_comp(struct sa_defrag_extent_backref
*b1
,
2207 struct sa_defrag_extent_backref
*b2
)
2209 if (b1
->root_id
< b2
->root_id
)
2211 else if (b1
->root_id
> b2
->root_id
)
2214 if (b1
->inum
< b2
->inum
)
2216 else if (b1
->inum
> b2
->inum
)
2219 if (b1
->file_pos
< b2
->file_pos
)
2221 else if (b1
->file_pos
> b2
->file_pos
)
2225 * [------------------------------] ===> (a range of space)
2226 * |<--->| |<---->| =============> (fs/file tree A)
2227 * |<---------------------------->| ===> (fs/file tree B)
2229 * A range of space can refer to two file extents in one tree while
2230 * refer to only one file extent in another tree.
2232 * So we may process a disk offset more than one time(two extents in A)
2233 * and locate at the same extent(one extent in B), then insert two same
2234 * backrefs(both refer to the extent in B).
2239 static void backref_insert(struct rb_root
*root
,
2240 struct sa_defrag_extent_backref
*backref
)
2242 struct rb_node
**p
= &root
->rb_node
;
2243 struct rb_node
*parent
= NULL
;
2244 struct sa_defrag_extent_backref
*entry
;
2249 entry
= rb_entry(parent
, struct sa_defrag_extent_backref
, node
);
2251 ret
= backref_comp(backref
, entry
);
2255 p
= &(*p
)->rb_right
;
2258 rb_link_node(&backref
->node
, parent
, p
);
2259 rb_insert_color(&backref
->node
, root
);
2263 * Note the backref might has changed, and in this case we just return 0.
2265 static noinline
int record_one_backref(u64 inum
, u64 offset
, u64 root_id
,
2268 struct btrfs_file_extent_item
*extent
;
2269 struct btrfs_fs_info
*fs_info
;
2270 struct old_sa_defrag_extent
*old
= ctx
;
2271 struct new_sa_defrag_extent
*new = old
->new;
2272 struct btrfs_path
*path
= new->path
;
2273 struct btrfs_key key
;
2274 struct btrfs_root
*root
;
2275 struct sa_defrag_extent_backref
*backref
;
2276 struct extent_buffer
*leaf
;
2277 struct inode
*inode
= new->inode
;
2283 if (BTRFS_I(inode
)->root
->root_key
.objectid
== root_id
&&
2284 inum
== btrfs_ino(inode
))
2287 key
.objectid
= root_id
;
2288 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2289 key
.offset
= (u64
)-1;
2291 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2292 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2294 if (PTR_ERR(root
) == -ENOENT
)
2297 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2298 inum
, offset
, root_id
);
2299 return PTR_ERR(root
);
2302 key
.objectid
= inum
;
2303 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2304 if (offset
> (u64
)-1 << 32)
2307 key
.offset
= offset
;
2309 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2310 if (WARN_ON(ret
< 0))
2317 leaf
= path
->nodes
[0];
2318 slot
= path
->slots
[0];
2320 if (slot
>= btrfs_header_nritems(leaf
)) {
2321 ret
= btrfs_next_leaf(root
, path
);
2324 } else if (ret
> 0) {
2333 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2335 if (key
.objectid
> inum
)
2338 if (key
.objectid
< inum
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2341 extent
= btrfs_item_ptr(leaf
, slot
,
2342 struct btrfs_file_extent_item
);
2344 if (btrfs_file_extent_disk_bytenr(leaf
, extent
) != old
->bytenr
)
2348 * 'offset' refers to the exact key.offset,
2349 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2350 * (key.offset - extent_offset).
2352 if (key
.offset
!= offset
)
2355 extent_offset
= btrfs_file_extent_offset(leaf
, extent
);
2356 num_bytes
= btrfs_file_extent_num_bytes(leaf
, extent
);
2358 if (extent_offset
>= old
->extent_offset
+ old
->offset
+
2359 old
->len
|| extent_offset
+ num_bytes
<=
2360 old
->extent_offset
+ old
->offset
)
2365 backref
= kmalloc(sizeof(*backref
), GFP_NOFS
);
2371 backref
->root_id
= root_id
;
2372 backref
->inum
= inum
;
2373 backref
->file_pos
= offset
;
2374 backref
->num_bytes
= num_bytes
;
2375 backref
->extent_offset
= extent_offset
;
2376 backref
->generation
= btrfs_file_extent_generation(leaf
, extent
);
2378 backref_insert(&new->root
, backref
);
2381 btrfs_release_path(path
);
2386 static noinline
bool record_extent_backrefs(struct btrfs_path
*path
,
2387 struct new_sa_defrag_extent
*new)
2389 struct btrfs_fs_info
*fs_info
= BTRFS_I(new->inode
)->root
->fs_info
;
2390 struct old_sa_defrag_extent
*old
, *tmp
;
2395 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2396 ret
= iterate_inodes_from_logical(old
->bytenr
+
2397 old
->extent_offset
, fs_info
,
2398 path
, record_one_backref
,
2400 if (ret
< 0 && ret
!= -ENOENT
)
2403 /* no backref to be processed for this extent */
2405 list_del(&old
->list
);
2410 if (list_empty(&new->head
))
2416 static int relink_is_mergable(struct extent_buffer
*leaf
,
2417 struct btrfs_file_extent_item
*fi
,
2418 struct new_sa_defrag_extent
*new)
2420 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) != new->bytenr
)
2423 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
2426 if (btrfs_file_extent_compression(leaf
, fi
) != new->compress_type
)
2429 if (btrfs_file_extent_encryption(leaf
, fi
) ||
2430 btrfs_file_extent_other_encoding(leaf
, fi
))
2437 * Note the backref might has changed, and in this case we just return 0.
2439 static noinline
int relink_extent_backref(struct btrfs_path
*path
,
2440 struct sa_defrag_extent_backref
*prev
,
2441 struct sa_defrag_extent_backref
*backref
)
2443 struct btrfs_file_extent_item
*extent
;
2444 struct btrfs_file_extent_item
*item
;
2445 struct btrfs_ordered_extent
*ordered
;
2446 struct btrfs_trans_handle
*trans
;
2447 struct btrfs_fs_info
*fs_info
;
2448 struct btrfs_root
*root
;
2449 struct btrfs_key key
;
2450 struct extent_buffer
*leaf
;
2451 struct old_sa_defrag_extent
*old
= backref
->old
;
2452 struct new_sa_defrag_extent
*new = old
->new;
2453 struct inode
*src_inode
= new->inode
;
2454 struct inode
*inode
;
2455 struct extent_state
*cached
= NULL
;
2464 if (prev
&& prev
->root_id
== backref
->root_id
&&
2465 prev
->inum
== backref
->inum
&&
2466 prev
->file_pos
+ prev
->num_bytes
== backref
->file_pos
)
2469 /* step 1: get root */
2470 key
.objectid
= backref
->root_id
;
2471 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2472 key
.offset
= (u64
)-1;
2474 fs_info
= BTRFS_I(src_inode
)->root
->fs_info
;
2475 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
2477 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2479 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2480 if (PTR_ERR(root
) == -ENOENT
)
2482 return PTR_ERR(root
);
2485 if (btrfs_root_readonly(root
)) {
2486 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2490 /* step 2: get inode */
2491 key
.objectid
= backref
->inum
;
2492 key
.type
= BTRFS_INODE_ITEM_KEY
;
2495 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
2496 if (IS_ERR(inode
)) {
2497 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2501 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2503 /* step 3: relink backref */
2504 lock_start
= backref
->file_pos
;
2505 lock_end
= backref
->file_pos
+ backref
->num_bytes
- 1;
2506 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2509 ordered
= btrfs_lookup_first_ordered_extent(inode
, lock_end
);
2511 btrfs_put_ordered_extent(ordered
);
2515 trans
= btrfs_join_transaction(root
);
2516 if (IS_ERR(trans
)) {
2517 ret
= PTR_ERR(trans
);
2521 key
.objectid
= backref
->inum
;
2522 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2523 key
.offset
= backref
->file_pos
;
2525 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2528 } else if (ret
> 0) {
2533 extent
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2534 struct btrfs_file_extent_item
);
2536 if (btrfs_file_extent_generation(path
->nodes
[0], extent
) !=
2537 backref
->generation
)
2540 btrfs_release_path(path
);
2542 start
= backref
->file_pos
;
2543 if (backref
->extent_offset
< old
->extent_offset
+ old
->offset
)
2544 start
+= old
->extent_offset
+ old
->offset
-
2545 backref
->extent_offset
;
2547 len
= min(backref
->extent_offset
+ backref
->num_bytes
,
2548 old
->extent_offset
+ old
->offset
+ old
->len
);
2549 len
-= max(backref
->extent_offset
, old
->extent_offset
+ old
->offset
);
2551 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
2556 key
.objectid
= btrfs_ino(inode
);
2557 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2560 path
->leave_spinning
= 1;
2562 struct btrfs_file_extent_item
*fi
;
2564 struct btrfs_key found_key
;
2566 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2571 leaf
= path
->nodes
[0];
2572 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2574 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2575 struct btrfs_file_extent_item
);
2576 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
2578 if (extent_len
+ found_key
.offset
== start
&&
2579 relink_is_mergable(leaf
, fi
, new)) {
2580 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2582 btrfs_mark_buffer_dirty(leaf
);
2583 inode_add_bytes(inode
, len
);
2589 btrfs_release_path(path
);
2594 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2597 btrfs_abort_transaction(trans
, root
, ret
);
2601 leaf
= path
->nodes
[0];
2602 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2603 struct btrfs_file_extent_item
);
2604 btrfs_set_file_extent_disk_bytenr(leaf
, item
, new->bytenr
);
2605 btrfs_set_file_extent_disk_num_bytes(leaf
, item
, new->disk_len
);
2606 btrfs_set_file_extent_offset(leaf
, item
, start
- new->file_pos
);
2607 btrfs_set_file_extent_num_bytes(leaf
, item
, len
);
2608 btrfs_set_file_extent_ram_bytes(leaf
, item
, new->len
);
2609 btrfs_set_file_extent_generation(leaf
, item
, trans
->transid
);
2610 btrfs_set_file_extent_type(leaf
, item
, BTRFS_FILE_EXTENT_REG
);
2611 btrfs_set_file_extent_compression(leaf
, item
, new->compress_type
);
2612 btrfs_set_file_extent_encryption(leaf
, item
, 0);
2613 btrfs_set_file_extent_other_encoding(leaf
, item
, 0);
2615 btrfs_mark_buffer_dirty(leaf
);
2616 inode_add_bytes(inode
, len
);
2617 btrfs_release_path(path
);
2619 ret
= btrfs_inc_extent_ref(trans
, root
, new->bytenr
,
2621 backref
->root_id
, backref
->inum
,
2622 new->file_pos
); /* start - extent_offset */
2624 btrfs_abort_transaction(trans
, root
, ret
);
2630 btrfs_release_path(path
);
2631 path
->leave_spinning
= 0;
2632 btrfs_end_transaction(trans
, root
);
2634 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2640 static void free_sa_defrag_extent(struct new_sa_defrag_extent
*new)
2642 struct old_sa_defrag_extent
*old
, *tmp
;
2647 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2653 static void relink_file_extents(struct new_sa_defrag_extent
*new)
2655 struct btrfs_path
*path
;
2656 struct sa_defrag_extent_backref
*backref
;
2657 struct sa_defrag_extent_backref
*prev
= NULL
;
2658 struct inode
*inode
;
2659 struct btrfs_root
*root
;
2660 struct rb_node
*node
;
2664 root
= BTRFS_I(inode
)->root
;
2666 path
= btrfs_alloc_path();
2670 if (!record_extent_backrefs(path
, new)) {
2671 btrfs_free_path(path
);
2674 btrfs_release_path(path
);
2677 node
= rb_first(&new->root
);
2680 rb_erase(node
, &new->root
);
2682 backref
= rb_entry(node
, struct sa_defrag_extent_backref
, node
);
2684 ret
= relink_extent_backref(path
, prev
, backref
);
2697 btrfs_free_path(path
);
2699 free_sa_defrag_extent(new);
2701 atomic_dec(&root
->fs_info
->defrag_running
);
2702 wake_up(&root
->fs_info
->transaction_wait
);
2705 static struct new_sa_defrag_extent
*
2706 record_old_file_extents(struct inode
*inode
,
2707 struct btrfs_ordered_extent
*ordered
)
2709 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2710 struct btrfs_path
*path
;
2711 struct btrfs_key key
;
2712 struct old_sa_defrag_extent
*old
;
2713 struct new_sa_defrag_extent
*new;
2716 new = kmalloc(sizeof(*new), GFP_NOFS
);
2721 new->file_pos
= ordered
->file_offset
;
2722 new->len
= ordered
->len
;
2723 new->bytenr
= ordered
->start
;
2724 new->disk_len
= ordered
->disk_len
;
2725 new->compress_type
= ordered
->compress_type
;
2726 new->root
= RB_ROOT
;
2727 INIT_LIST_HEAD(&new->head
);
2729 path
= btrfs_alloc_path();
2733 key
.objectid
= btrfs_ino(inode
);
2734 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2735 key
.offset
= new->file_pos
;
2737 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2740 if (ret
> 0 && path
->slots
[0] > 0)
2743 /* find out all the old extents for the file range */
2745 struct btrfs_file_extent_item
*extent
;
2746 struct extent_buffer
*l
;
2755 slot
= path
->slots
[0];
2757 if (slot
>= btrfs_header_nritems(l
)) {
2758 ret
= btrfs_next_leaf(root
, path
);
2766 btrfs_item_key_to_cpu(l
, &key
, slot
);
2768 if (key
.objectid
!= btrfs_ino(inode
))
2770 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2772 if (key
.offset
>= new->file_pos
+ new->len
)
2775 extent
= btrfs_item_ptr(l
, slot
, struct btrfs_file_extent_item
);
2777 num_bytes
= btrfs_file_extent_num_bytes(l
, extent
);
2778 if (key
.offset
+ num_bytes
< new->file_pos
)
2781 disk_bytenr
= btrfs_file_extent_disk_bytenr(l
, extent
);
2785 extent_offset
= btrfs_file_extent_offset(l
, extent
);
2787 old
= kmalloc(sizeof(*old
), GFP_NOFS
);
2791 offset
= max(new->file_pos
, key
.offset
);
2792 end
= min(new->file_pos
+ new->len
, key
.offset
+ num_bytes
);
2794 old
->bytenr
= disk_bytenr
;
2795 old
->extent_offset
= extent_offset
;
2796 old
->offset
= offset
- key
.offset
;
2797 old
->len
= end
- offset
;
2800 list_add_tail(&old
->list
, &new->head
);
2806 btrfs_free_path(path
);
2807 atomic_inc(&root
->fs_info
->defrag_running
);
2812 btrfs_free_path(path
);
2814 free_sa_defrag_extent(new);
2818 static void btrfs_release_delalloc_bytes(struct btrfs_root
*root
,
2821 struct btrfs_block_group_cache
*cache
;
2823 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
2826 spin_lock(&cache
->lock
);
2827 cache
->delalloc_bytes
-= len
;
2828 spin_unlock(&cache
->lock
);
2830 btrfs_put_block_group(cache
);
2833 /* as ordered data IO finishes, this gets called so we can finish
2834 * an ordered extent if the range of bytes in the file it covers are
2837 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
)
2839 struct inode
*inode
= ordered_extent
->inode
;
2840 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2841 struct btrfs_trans_handle
*trans
= NULL
;
2842 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2843 struct extent_state
*cached_state
= NULL
;
2844 struct new_sa_defrag_extent
*new = NULL
;
2845 int compress_type
= 0;
2847 u64 logical_len
= ordered_extent
->len
;
2849 bool truncated
= false;
2851 nolock
= btrfs_is_free_space_inode(inode
);
2853 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
2858 btrfs_free_io_failure_record(inode
, ordered_extent
->file_offset
,
2859 ordered_extent
->file_offset
+
2860 ordered_extent
->len
- 1);
2862 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered_extent
->flags
)) {
2864 logical_len
= ordered_extent
->truncated_len
;
2865 /* Truncated the entire extent, don't bother adding */
2870 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
2871 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
2874 * For mwrite(mmap + memset to write) case, we still reserve
2875 * space for NOCOW range.
2876 * As NOCOW won't cause a new delayed ref, just free the space
2878 btrfs_qgroup_free_data(inode
, ordered_extent
->file_offset
,
2879 ordered_extent
->len
);
2880 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2882 trans
= btrfs_join_transaction_nolock(root
);
2884 trans
= btrfs_join_transaction(root
);
2885 if (IS_ERR(trans
)) {
2886 ret
= PTR_ERR(trans
);
2890 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2891 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2892 if (ret
) /* -ENOMEM or corruption */
2893 btrfs_abort_transaction(trans
, root
, ret
);
2897 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
2898 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2901 ret
= test_range_bit(io_tree
, ordered_extent
->file_offset
,
2902 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2903 EXTENT_DEFRAG
, 1, cached_state
);
2905 u64 last_snapshot
= btrfs_root_last_snapshot(&root
->root_item
);
2906 if (0 && last_snapshot
>= BTRFS_I(inode
)->generation
)
2907 /* the inode is shared */
2908 new = record_old_file_extents(inode
, ordered_extent
);
2910 clear_extent_bit(io_tree
, ordered_extent
->file_offset
,
2911 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2912 EXTENT_DEFRAG
, 0, 0, &cached_state
, GFP_NOFS
);
2916 trans
= btrfs_join_transaction_nolock(root
);
2918 trans
= btrfs_join_transaction(root
);
2919 if (IS_ERR(trans
)) {
2920 ret
= PTR_ERR(trans
);
2925 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2927 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
2928 compress_type
= ordered_extent
->compress_type
;
2929 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
2930 BUG_ON(compress_type
);
2931 ret
= btrfs_mark_extent_written(trans
, inode
,
2932 ordered_extent
->file_offset
,
2933 ordered_extent
->file_offset
+
2936 BUG_ON(root
== root
->fs_info
->tree_root
);
2937 ret
= insert_reserved_file_extent(trans
, inode
,
2938 ordered_extent
->file_offset
,
2939 ordered_extent
->start
,
2940 ordered_extent
->disk_len
,
2941 logical_len
, logical_len
,
2942 compress_type
, 0, 0,
2943 BTRFS_FILE_EXTENT_REG
);
2945 btrfs_release_delalloc_bytes(root
,
2946 ordered_extent
->start
,
2947 ordered_extent
->disk_len
);
2949 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
2950 ordered_extent
->file_offset
, ordered_extent
->len
,
2953 btrfs_abort_transaction(trans
, root
, ret
);
2957 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
2958 &ordered_extent
->list
);
2960 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2961 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2962 if (ret
) { /* -ENOMEM or corruption */
2963 btrfs_abort_transaction(trans
, root
, ret
);
2968 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
2969 ordered_extent
->file_offset
+
2970 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
2972 if (root
!= root
->fs_info
->tree_root
)
2973 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
2975 btrfs_end_transaction(trans
, root
);
2977 if (ret
|| truncated
) {
2981 start
= ordered_extent
->file_offset
+ logical_len
;
2983 start
= ordered_extent
->file_offset
;
2984 end
= ordered_extent
->file_offset
+ ordered_extent
->len
- 1;
2985 clear_extent_uptodate(io_tree
, start
, end
, NULL
, GFP_NOFS
);
2987 /* Drop the cache for the part of the extent we didn't write. */
2988 btrfs_drop_extent_cache(inode
, start
, end
, 0);
2991 * If the ordered extent had an IOERR or something else went
2992 * wrong we need to return the space for this ordered extent
2993 * back to the allocator. We only free the extent in the
2994 * truncated case if we didn't write out the extent at all.
2996 if ((ret
|| !logical_len
) &&
2997 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
2998 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
))
2999 btrfs_free_reserved_extent(root
, ordered_extent
->start
,
3000 ordered_extent
->disk_len
, 1);
3005 * This needs to be done to make sure anybody waiting knows we are done
3006 * updating everything for this ordered extent.
3008 btrfs_remove_ordered_extent(inode
, ordered_extent
);
3010 /* for snapshot-aware defrag */
3013 free_sa_defrag_extent(new);
3014 atomic_dec(&root
->fs_info
->defrag_running
);
3016 relink_file_extents(new);
3021 btrfs_put_ordered_extent(ordered_extent
);
3022 /* once for the tree */
3023 btrfs_put_ordered_extent(ordered_extent
);
3028 static void finish_ordered_fn(struct btrfs_work
*work
)
3030 struct btrfs_ordered_extent
*ordered_extent
;
3031 ordered_extent
= container_of(work
, struct btrfs_ordered_extent
, work
);
3032 btrfs_finish_ordered_io(ordered_extent
);
3035 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
3036 struct extent_state
*state
, int uptodate
)
3038 struct inode
*inode
= page
->mapping
->host
;
3039 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3040 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
3041 struct btrfs_workqueue
*wq
;
3042 btrfs_work_func_t func
;
3044 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
3046 ClearPagePrivate2(page
);
3047 if (!btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
3048 end
- start
+ 1, uptodate
))
3051 if (btrfs_is_free_space_inode(inode
)) {
3052 wq
= root
->fs_info
->endio_freespace_worker
;
3053 func
= btrfs_freespace_write_helper
;
3055 wq
= root
->fs_info
->endio_write_workers
;
3056 func
= btrfs_endio_write_helper
;
3059 btrfs_init_work(&ordered_extent
->work
, func
, finish_ordered_fn
, NULL
,
3061 btrfs_queue_work(wq
, &ordered_extent
->work
);
3066 static int __readpage_endio_check(struct inode
*inode
,
3067 struct btrfs_io_bio
*io_bio
,
3068 int icsum
, struct page
*page
,
3069 int pgoff
, u64 start
, size_t len
)
3075 csum_expected
= *(((u32
*)io_bio
->csum
) + icsum
);
3077 kaddr
= kmap_atomic(page
);
3078 csum
= btrfs_csum_data(kaddr
+ pgoff
, csum
, len
);
3079 btrfs_csum_final(csum
, (char *)&csum
);
3080 if (csum
!= csum_expected
)
3083 kunmap_atomic(kaddr
);
3086 btrfs_warn_rl(BTRFS_I(inode
)->root
->fs_info
,
3087 "csum failed ino %llu off %llu csum %u expected csum %u",
3088 btrfs_ino(inode
), start
, csum
, csum_expected
);
3089 memset(kaddr
+ pgoff
, 1, len
);
3090 flush_dcache_page(page
);
3091 kunmap_atomic(kaddr
);
3092 if (csum_expected
== 0)
3098 * when reads are done, we need to check csums to verify the data is correct
3099 * if there's a match, we allow the bio to finish. If not, the code in
3100 * extent_io.c will try to find good copies for us.
3102 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
3103 u64 phy_offset
, struct page
*page
,
3104 u64 start
, u64 end
, int mirror
)
3106 size_t offset
= start
- page_offset(page
);
3107 struct inode
*inode
= page
->mapping
->host
;
3108 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3109 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3111 if (PageChecked(page
)) {
3112 ClearPageChecked(page
);
3116 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
3119 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
3120 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
3121 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
);
3125 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
3126 return __readpage_endio_check(inode
, io_bio
, phy_offset
, page
, offset
,
3127 start
, (size_t)(end
- start
+ 1));
3130 void btrfs_add_delayed_iput(struct inode
*inode
)
3132 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
3133 struct btrfs_inode
*binode
= BTRFS_I(inode
);
3135 if (atomic_add_unless(&inode
->i_count
, -1, 1))
3138 spin_lock(&fs_info
->delayed_iput_lock
);
3139 if (binode
->delayed_iput_count
== 0) {
3140 ASSERT(list_empty(&binode
->delayed_iput
));
3141 list_add_tail(&binode
->delayed_iput
, &fs_info
->delayed_iputs
);
3143 binode
->delayed_iput_count
++;
3145 spin_unlock(&fs_info
->delayed_iput_lock
);
3148 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
3150 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3152 spin_lock(&fs_info
->delayed_iput_lock
);
3153 while (!list_empty(&fs_info
->delayed_iputs
)) {
3154 struct btrfs_inode
*inode
;
3156 inode
= list_first_entry(&fs_info
->delayed_iputs
,
3157 struct btrfs_inode
, delayed_iput
);
3158 if (inode
->delayed_iput_count
) {
3159 inode
->delayed_iput_count
--;
3160 list_move_tail(&inode
->delayed_iput
,
3161 &fs_info
->delayed_iputs
);
3163 list_del_init(&inode
->delayed_iput
);
3165 spin_unlock(&fs_info
->delayed_iput_lock
);
3166 iput(&inode
->vfs_inode
);
3167 spin_lock(&fs_info
->delayed_iput_lock
);
3169 spin_unlock(&fs_info
->delayed_iput_lock
);
3173 * This is called in transaction commit time. If there are no orphan
3174 * files in the subvolume, it removes orphan item and frees block_rsv
3177 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
3178 struct btrfs_root
*root
)
3180 struct btrfs_block_rsv
*block_rsv
;
3183 if (atomic_read(&root
->orphan_inodes
) ||
3184 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
3187 spin_lock(&root
->orphan_lock
);
3188 if (atomic_read(&root
->orphan_inodes
)) {
3189 spin_unlock(&root
->orphan_lock
);
3193 if (root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
) {
3194 spin_unlock(&root
->orphan_lock
);
3198 block_rsv
= root
->orphan_block_rsv
;
3199 root
->orphan_block_rsv
= NULL
;
3200 spin_unlock(&root
->orphan_lock
);
3202 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
) &&
3203 btrfs_root_refs(&root
->root_item
) > 0) {
3204 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
3205 root
->root_key
.objectid
);
3207 btrfs_abort_transaction(trans
, root
, ret
);
3209 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
,
3214 WARN_ON(block_rsv
->size
> 0);
3215 btrfs_free_block_rsv(root
, block_rsv
);
3220 * This creates an orphan entry for the given inode in case something goes
3221 * wrong in the middle of an unlink/truncate.
3223 * NOTE: caller of this function should reserve 5 units of metadata for
3226 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
3228 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3229 struct btrfs_block_rsv
*block_rsv
= NULL
;
3234 if (!root
->orphan_block_rsv
) {
3235 block_rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
3240 spin_lock(&root
->orphan_lock
);
3241 if (!root
->orphan_block_rsv
) {
3242 root
->orphan_block_rsv
= block_rsv
;
3243 } else if (block_rsv
) {
3244 btrfs_free_block_rsv(root
, block_rsv
);
3248 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3249 &BTRFS_I(inode
)->runtime_flags
)) {
3252 * For proper ENOSPC handling, we should do orphan
3253 * cleanup when mounting. But this introduces backward
3254 * compatibility issue.
3256 if (!xchg(&root
->orphan_item_inserted
, 1))
3262 atomic_inc(&root
->orphan_inodes
);
3265 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3266 &BTRFS_I(inode
)->runtime_flags
))
3268 spin_unlock(&root
->orphan_lock
);
3270 /* grab metadata reservation from transaction handle */
3272 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
3275 atomic_dec(&root
->orphan_inodes
);
3276 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3277 &BTRFS_I(inode
)->runtime_flags
);
3279 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3280 &BTRFS_I(inode
)->runtime_flags
);
3285 /* insert an orphan item to track this unlinked/truncated file */
3287 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
3289 atomic_dec(&root
->orphan_inodes
);
3291 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3292 &BTRFS_I(inode
)->runtime_flags
);
3293 btrfs_orphan_release_metadata(inode
);
3295 if (ret
!= -EEXIST
) {
3296 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3297 &BTRFS_I(inode
)->runtime_flags
);
3298 btrfs_abort_transaction(trans
, root
, ret
);
3305 /* insert an orphan item to track subvolume contains orphan files */
3307 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
3308 root
->root_key
.objectid
);
3309 if (ret
&& ret
!= -EEXIST
) {
3310 btrfs_abort_transaction(trans
, root
, ret
);
3318 * We have done the truncate/delete so we can go ahead and remove the orphan
3319 * item for this particular inode.
3321 static int btrfs_orphan_del(struct btrfs_trans_handle
*trans
,
3322 struct inode
*inode
)
3324 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3325 int delete_item
= 0;
3326 int release_rsv
= 0;
3329 spin_lock(&root
->orphan_lock
);
3330 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3331 &BTRFS_I(inode
)->runtime_flags
))
3334 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3335 &BTRFS_I(inode
)->runtime_flags
))
3337 spin_unlock(&root
->orphan_lock
);
3340 atomic_dec(&root
->orphan_inodes
);
3342 ret
= btrfs_del_orphan_item(trans
, root
,
3347 btrfs_orphan_release_metadata(inode
);
3353 * this cleans up any orphans that may be left on the list from the last use
3356 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
3358 struct btrfs_path
*path
;
3359 struct extent_buffer
*leaf
;
3360 struct btrfs_key key
, found_key
;
3361 struct btrfs_trans_handle
*trans
;
3362 struct inode
*inode
;
3363 u64 last_objectid
= 0;
3364 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
3366 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
3369 path
= btrfs_alloc_path();
3374 path
->reada
= READA_BACK
;
3376 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
3377 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
3378 key
.offset
= (u64
)-1;
3381 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3386 * if ret == 0 means we found what we were searching for, which
3387 * is weird, but possible, so only screw with path if we didn't
3388 * find the key and see if we have stuff that matches
3392 if (path
->slots
[0] == 0)
3397 /* pull out the item */
3398 leaf
= path
->nodes
[0];
3399 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3401 /* make sure the item matches what we want */
3402 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
3404 if (found_key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
3407 /* release the path since we're done with it */
3408 btrfs_release_path(path
);
3411 * this is where we are basically btrfs_lookup, without the
3412 * crossing root thing. we store the inode number in the
3413 * offset of the orphan item.
3416 if (found_key
.offset
== last_objectid
) {
3417 btrfs_err(root
->fs_info
,
3418 "Error removing orphan entry, stopping orphan cleanup");
3423 last_objectid
= found_key
.offset
;
3425 found_key
.objectid
= found_key
.offset
;
3426 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
3427 found_key
.offset
= 0;
3428 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
3429 ret
= PTR_ERR_OR_ZERO(inode
);
3430 if (ret
&& ret
!= -ESTALE
)
3433 if (ret
== -ESTALE
&& root
== root
->fs_info
->tree_root
) {
3434 struct btrfs_root
*dead_root
;
3435 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3436 int is_dead_root
= 0;
3439 * this is an orphan in the tree root. Currently these
3440 * could come from 2 sources:
3441 * a) a snapshot deletion in progress
3442 * b) a free space cache inode
3443 * We need to distinguish those two, as the snapshot
3444 * orphan must not get deleted.
3445 * find_dead_roots already ran before us, so if this
3446 * is a snapshot deletion, we should find the root
3447 * in the dead_roots list
3449 spin_lock(&fs_info
->trans_lock
);
3450 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
3452 if (dead_root
->root_key
.objectid
==
3453 found_key
.objectid
) {
3458 spin_unlock(&fs_info
->trans_lock
);
3460 /* prevent this orphan from being found again */
3461 key
.offset
= found_key
.objectid
- 1;
3466 * Inode is already gone but the orphan item is still there,
3467 * kill the orphan item.
3469 if (ret
== -ESTALE
) {
3470 trans
= btrfs_start_transaction(root
, 1);
3471 if (IS_ERR(trans
)) {
3472 ret
= PTR_ERR(trans
);
3475 btrfs_debug(root
->fs_info
, "auto deleting %Lu",
3476 found_key
.objectid
);
3477 ret
= btrfs_del_orphan_item(trans
, root
,
3478 found_key
.objectid
);
3479 btrfs_end_transaction(trans
, root
);
3486 * add this inode to the orphan list so btrfs_orphan_del does
3487 * the proper thing when we hit it
3489 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3490 &BTRFS_I(inode
)->runtime_flags
);
3491 atomic_inc(&root
->orphan_inodes
);
3493 /* if we have links, this was a truncate, lets do that */
3494 if (inode
->i_nlink
) {
3495 if (WARN_ON(!S_ISREG(inode
->i_mode
))) {
3501 /* 1 for the orphan item deletion. */
3502 trans
= btrfs_start_transaction(root
, 1);
3503 if (IS_ERR(trans
)) {
3505 ret
= PTR_ERR(trans
);
3508 ret
= btrfs_orphan_add(trans
, inode
);
3509 btrfs_end_transaction(trans
, root
);
3515 ret
= btrfs_truncate(inode
);
3517 btrfs_orphan_del(NULL
, inode
);
3522 /* this will do delete_inode and everything for us */
3527 /* release the path since we're done with it */
3528 btrfs_release_path(path
);
3530 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
3532 if (root
->orphan_block_rsv
)
3533 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
3536 if (root
->orphan_block_rsv
||
3537 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
)) {
3538 trans
= btrfs_join_transaction(root
);
3540 btrfs_end_transaction(trans
, root
);
3544 btrfs_debug(root
->fs_info
, "unlinked %d orphans", nr_unlink
);
3546 btrfs_debug(root
->fs_info
, "truncated %d orphans", nr_truncate
);
3550 btrfs_err(root
->fs_info
,
3551 "could not do orphan cleanup %d", ret
);
3552 btrfs_free_path(path
);
3557 * very simple check to peek ahead in the leaf looking for xattrs. If we
3558 * don't find any xattrs, we know there can't be any acls.
3560 * slot is the slot the inode is in, objectid is the objectid of the inode
3562 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
3563 int slot
, u64 objectid
,
3564 int *first_xattr_slot
)
3566 u32 nritems
= btrfs_header_nritems(leaf
);
3567 struct btrfs_key found_key
;
3568 static u64 xattr_access
= 0;
3569 static u64 xattr_default
= 0;
3572 if (!xattr_access
) {
3573 xattr_access
= btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS
,
3574 strlen(XATTR_NAME_POSIX_ACL_ACCESS
));
3575 xattr_default
= btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT
,
3576 strlen(XATTR_NAME_POSIX_ACL_DEFAULT
));
3580 *first_xattr_slot
= -1;
3581 while (slot
< nritems
) {
3582 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3584 /* we found a different objectid, there must not be acls */
3585 if (found_key
.objectid
!= objectid
)
3588 /* we found an xattr, assume we've got an acl */
3589 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
) {
3590 if (*first_xattr_slot
== -1)
3591 *first_xattr_slot
= slot
;
3592 if (found_key
.offset
== xattr_access
||
3593 found_key
.offset
== xattr_default
)
3598 * we found a key greater than an xattr key, there can't
3599 * be any acls later on
3601 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
3608 * it goes inode, inode backrefs, xattrs, extents,
3609 * so if there are a ton of hard links to an inode there can
3610 * be a lot of backrefs. Don't waste time searching too hard,
3611 * this is just an optimization
3616 /* we hit the end of the leaf before we found an xattr or
3617 * something larger than an xattr. We have to assume the inode
3620 if (*first_xattr_slot
== -1)
3621 *first_xattr_slot
= slot
;
3626 * read an inode from the btree into the in-memory inode
3628 static void btrfs_read_locked_inode(struct inode
*inode
)
3630 struct btrfs_path
*path
;
3631 struct extent_buffer
*leaf
;
3632 struct btrfs_inode_item
*inode_item
;
3633 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3634 struct btrfs_key location
;
3639 bool filled
= false;
3640 int first_xattr_slot
;
3642 ret
= btrfs_fill_inode(inode
, &rdev
);
3646 path
= btrfs_alloc_path();
3650 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
3652 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
3656 leaf
= path
->nodes
[0];
3661 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3662 struct btrfs_inode_item
);
3663 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
3664 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
3665 i_uid_write(inode
, btrfs_inode_uid(leaf
, inode_item
));
3666 i_gid_write(inode
, btrfs_inode_gid(leaf
, inode_item
));
3667 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
3669 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->atime
);
3670 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->atime
);
3672 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->mtime
);
3673 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->mtime
);
3675 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->ctime
);
3676 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->ctime
);
3678 BTRFS_I(inode
)->i_otime
.tv_sec
=
3679 btrfs_timespec_sec(leaf
, &inode_item
->otime
);
3680 BTRFS_I(inode
)->i_otime
.tv_nsec
=
3681 btrfs_timespec_nsec(leaf
, &inode_item
->otime
);
3683 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
3684 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
3685 BTRFS_I(inode
)->last_trans
= btrfs_inode_transid(leaf
, inode_item
);
3687 inode
->i_version
= btrfs_inode_sequence(leaf
, inode_item
);
3688 inode
->i_generation
= BTRFS_I(inode
)->generation
;
3690 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
3692 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
3693 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
3697 * If we were modified in the current generation and evicted from memory
3698 * and then re-read we need to do a full sync since we don't have any
3699 * idea about which extents were modified before we were evicted from
3702 * This is required for both inode re-read from disk and delayed inode
3703 * in delayed_nodes_tree.
3705 if (BTRFS_I(inode
)->last_trans
== root
->fs_info
->generation
)
3706 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3707 &BTRFS_I(inode
)->runtime_flags
);
3710 * We don't persist the id of the transaction where an unlink operation
3711 * against the inode was last made. So here we assume the inode might
3712 * have been evicted, and therefore the exact value of last_unlink_trans
3713 * lost, and set it to last_trans to avoid metadata inconsistencies
3714 * between the inode and its parent if the inode is fsync'ed and the log
3715 * replayed. For example, in the scenario:
3718 * ln mydir/foo mydir/bar
3721 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3722 * xfs_io -c fsync mydir/foo
3724 * mount fs, triggers fsync log replay
3726 * We must make sure that when we fsync our inode foo we also log its
3727 * parent inode, otherwise after log replay the parent still has the
3728 * dentry with the "bar" name but our inode foo has a link count of 1
3729 * and doesn't have an inode ref with the name "bar" anymore.
3731 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3732 * but it guarantees correctness at the expense of occasional full
3733 * transaction commits on fsync if our inode is a directory, or if our
3734 * inode is not a directory, logging its parent unnecessarily.
3736 BTRFS_I(inode
)->last_unlink_trans
= BTRFS_I(inode
)->last_trans
;
3739 if (inode
->i_nlink
!= 1 ||
3740 path
->slots
[0] >= btrfs_header_nritems(leaf
))
3743 btrfs_item_key_to_cpu(leaf
, &location
, path
->slots
[0]);
3744 if (location
.objectid
!= btrfs_ino(inode
))
3747 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3748 if (location
.type
== BTRFS_INODE_REF_KEY
) {
3749 struct btrfs_inode_ref
*ref
;
3751 ref
= (struct btrfs_inode_ref
*)ptr
;
3752 BTRFS_I(inode
)->dir_index
= btrfs_inode_ref_index(leaf
, ref
);
3753 } else if (location
.type
== BTRFS_INODE_EXTREF_KEY
) {
3754 struct btrfs_inode_extref
*extref
;
3756 extref
= (struct btrfs_inode_extref
*)ptr
;
3757 BTRFS_I(inode
)->dir_index
= btrfs_inode_extref_index(leaf
,
3762 * try to precache a NULL acl entry for files that don't have
3763 * any xattrs or acls
3765 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
3766 btrfs_ino(inode
), &first_xattr_slot
);
3767 if (first_xattr_slot
!= -1) {
3768 path
->slots
[0] = first_xattr_slot
;
3769 ret
= btrfs_load_inode_props(inode
, path
);
3771 btrfs_err(root
->fs_info
,
3772 "error loading props for ino %llu (root %llu): %d",
3774 root
->root_key
.objectid
, ret
);
3776 btrfs_free_path(path
);
3779 cache_no_acl(inode
);
3781 switch (inode
->i_mode
& S_IFMT
) {
3783 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3784 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3785 inode
->i_fop
= &btrfs_file_operations
;
3786 inode
->i_op
= &btrfs_file_inode_operations
;
3789 inode
->i_fop
= &btrfs_dir_file_operations
;
3790 if (root
== root
->fs_info
->tree_root
)
3791 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
3793 inode
->i_op
= &btrfs_dir_inode_operations
;
3796 inode
->i_op
= &btrfs_symlink_inode_operations
;
3797 inode_nohighmem(inode
);
3798 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
3801 inode
->i_op
= &btrfs_special_inode_operations
;
3802 init_special_inode(inode
, inode
->i_mode
, rdev
);
3806 btrfs_update_iflags(inode
);
3810 btrfs_free_path(path
);
3811 make_bad_inode(inode
);
3815 * given a leaf and an inode, copy the inode fields into the leaf
3817 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
3818 struct extent_buffer
*leaf
,
3819 struct btrfs_inode_item
*item
,
3820 struct inode
*inode
)
3822 struct btrfs_map_token token
;
3824 btrfs_init_map_token(&token
);
3826 btrfs_set_token_inode_uid(leaf
, item
, i_uid_read(inode
), &token
);
3827 btrfs_set_token_inode_gid(leaf
, item
, i_gid_read(inode
), &token
);
3828 btrfs_set_token_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
,
3830 btrfs_set_token_inode_mode(leaf
, item
, inode
->i_mode
, &token
);
3831 btrfs_set_token_inode_nlink(leaf
, item
, inode
->i_nlink
, &token
);
3833 btrfs_set_token_timespec_sec(leaf
, &item
->atime
,
3834 inode
->i_atime
.tv_sec
, &token
);
3835 btrfs_set_token_timespec_nsec(leaf
, &item
->atime
,
3836 inode
->i_atime
.tv_nsec
, &token
);
3838 btrfs_set_token_timespec_sec(leaf
, &item
->mtime
,
3839 inode
->i_mtime
.tv_sec
, &token
);
3840 btrfs_set_token_timespec_nsec(leaf
, &item
->mtime
,
3841 inode
->i_mtime
.tv_nsec
, &token
);
3843 btrfs_set_token_timespec_sec(leaf
, &item
->ctime
,
3844 inode
->i_ctime
.tv_sec
, &token
);
3845 btrfs_set_token_timespec_nsec(leaf
, &item
->ctime
,
3846 inode
->i_ctime
.tv_nsec
, &token
);
3848 btrfs_set_token_timespec_sec(leaf
, &item
->otime
,
3849 BTRFS_I(inode
)->i_otime
.tv_sec
, &token
);
3850 btrfs_set_token_timespec_nsec(leaf
, &item
->otime
,
3851 BTRFS_I(inode
)->i_otime
.tv_nsec
, &token
);
3853 btrfs_set_token_inode_nbytes(leaf
, item
, inode_get_bytes(inode
),
3855 btrfs_set_token_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
,
3857 btrfs_set_token_inode_sequence(leaf
, item
, inode
->i_version
, &token
);
3858 btrfs_set_token_inode_transid(leaf
, item
, trans
->transid
, &token
);
3859 btrfs_set_token_inode_rdev(leaf
, item
, inode
->i_rdev
, &token
);
3860 btrfs_set_token_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
, &token
);
3861 btrfs_set_token_inode_block_group(leaf
, item
, 0, &token
);
3865 * copy everything in the in-memory inode into the btree.
3867 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
3868 struct btrfs_root
*root
, struct inode
*inode
)
3870 struct btrfs_inode_item
*inode_item
;
3871 struct btrfs_path
*path
;
3872 struct extent_buffer
*leaf
;
3875 path
= btrfs_alloc_path();
3879 path
->leave_spinning
= 1;
3880 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
3888 leaf
= path
->nodes
[0];
3889 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3890 struct btrfs_inode_item
);
3892 fill_inode_item(trans
, leaf
, inode_item
, inode
);
3893 btrfs_mark_buffer_dirty(leaf
);
3894 btrfs_set_inode_last_trans(trans
, inode
);
3897 btrfs_free_path(path
);
3902 * copy everything in the in-memory inode into the btree.
3904 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
3905 struct btrfs_root
*root
, struct inode
*inode
)
3910 * If the inode is a free space inode, we can deadlock during commit
3911 * if we put it into the delayed code.
3913 * The data relocation inode should also be directly updated
3916 if (!btrfs_is_free_space_inode(inode
)
3917 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
3918 && !root
->fs_info
->log_root_recovering
) {
3919 btrfs_update_root_times(trans
, root
);
3921 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
3923 btrfs_set_inode_last_trans(trans
, inode
);
3927 return btrfs_update_inode_item(trans
, root
, inode
);
3930 noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
3931 struct btrfs_root
*root
,
3932 struct inode
*inode
)
3936 ret
= btrfs_update_inode(trans
, root
, inode
);
3938 return btrfs_update_inode_item(trans
, root
, inode
);
3943 * unlink helper that gets used here in inode.c and in the tree logging
3944 * recovery code. It remove a link in a directory with a given name, and
3945 * also drops the back refs in the inode to the directory
3947 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
3948 struct btrfs_root
*root
,
3949 struct inode
*dir
, struct inode
*inode
,
3950 const char *name
, int name_len
)
3952 struct btrfs_path
*path
;
3954 struct extent_buffer
*leaf
;
3955 struct btrfs_dir_item
*di
;
3956 struct btrfs_key key
;
3958 u64 ino
= btrfs_ino(inode
);
3959 u64 dir_ino
= btrfs_ino(dir
);
3961 path
= btrfs_alloc_path();
3967 path
->leave_spinning
= 1;
3968 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3969 name
, name_len
, -1);
3978 leaf
= path
->nodes
[0];
3979 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3980 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3983 btrfs_release_path(path
);
3986 * If we don't have dir index, we have to get it by looking up
3987 * the inode ref, since we get the inode ref, remove it directly,
3988 * it is unnecessary to do delayed deletion.
3990 * But if we have dir index, needn't search inode ref to get it.
3991 * Since the inode ref is close to the inode item, it is better
3992 * that we delay to delete it, and just do this deletion when
3993 * we update the inode item.
3995 if (BTRFS_I(inode
)->dir_index
) {
3996 ret
= btrfs_delayed_delete_inode_ref(inode
);
3998 index
= BTRFS_I(inode
)->dir_index
;
4003 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
4006 btrfs_info(root
->fs_info
,
4007 "failed to delete reference to %.*s, inode %llu parent %llu",
4008 name_len
, name
, ino
, dir_ino
);
4009 btrfs_abort_transaction(trans
, root
, ret
);
4013 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
4015 btrfs_abort_transaction(trans
, root
, ret
);
4019 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
4021 if (ret
!= 0 && ret
!= -ENOENT
) {
4022 btrfs_abort_transaction(trans
, root
, ret
);
4026 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
4031 btrfs_abort_transaction(trans
, root
, ret
);
4033 btrfs_free_path(path
);
4037 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
4038 inode_inc_iversion(inode
);
4039 inode_inc_iversion(dir
);
4040 inode
->i_ctime
= dir
->i_mtime
=
4041 dir
->i_ctime
= current_fs_time(inode
->i_sb
);
4042 ret
= btrfs_update_inode(trans
, root
, dir
);
4047 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
4048 struct btrfs_root
*root
,
4049 struct inode
*dir
, struct inode
*inode
,
4050 const char *name
, int name_len
)
4053 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
4056 ret
= btrfs_update_inode(trans
, root
, inode
);
4062 * helper to start transaction for unlink and rmdir.
4064 * unlink and rmdir are special in btrfs, they do not always free space, so
4065 * if we cannot make our reservations the normal way try and see if there is
4066 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4067 * allow the unlink to occur.
4069 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
)
4071 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4074 * 1 for the possible orphan item
4075 * 1 for the dir item
4076 * 1 for the dir index
4077 * 1 for the inode ref
4080 return btrfs_start_transaction_fallback_global_rsv(root
, 5, 5);
4083 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
4085 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4086 struct btrfs_trans_handle
*trans
;
4087 struct inode
*inode
= d_inode(dentry
);
4090 trans
= __unlink_start_trans(dir
);
4092 return PTR_ERR(trans
);
4094 btrfs_record_unlink_dir(trans
, dir
, d_inode(dentry
), 0);
4096 ret
= btrfs_unlink_inode(trans
, root
, dir
, d_inode(dentry
),
4097 dentry
->d_name
.name
, dentry
->d_name
.len
);
4101 if (inode
->i_nlink
== 0) {
4102 ret
= btrfs_orphan_add(trans
, inode
);
4108 btrfs_end_transaction(trans
, root
);
4109 btrfs_btree_balance_dirty(root
);
4113 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
4114 struct btrfs_root
*root
,
4115 struct inode
*dir
, u64 objectid
,
4116 const char *name
, int name_len
)
4118 struct btrfs_path
*path
;
4119 struct extent_buffer
*leaf
;
4120 struct btrfs_dir_item
*di
;
4121 struct btrfs_key key
;
4124 u64 dir_ino
= btrfs_ino(dir
);
4126 path
= btrfs_alloc_path();
4130 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
4131 name
, name_len
, -1);
4132 if (IS_ERR_OR_NULL(di
)) {
4140 leaf
= path
->nodes
[0];
4141 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
4142 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
4143 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
4145 btrfs_abort_transaction(trans
, root
, ret
);
4148 btrfs_release_path(path
);
4150 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
4151 objectid
, root
->root_key
.objectid
,
4152 dir_ino
, &index
, name
, name_len
);
4154 if (ret
!= -ENOENT
) {
4155 btrfs_abort_transaction(trans
, root
, ret
);
4158 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
4160 if (IS_ERR_OR_NULL(di
)) {
4165 btrfs_abort_transaction(trans
, root
, ret
);
4169 leaf
= path
->nodes
[0];
4170 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4171 btrfs_release_path(path
);
4174 btrfs_release_path(path
);
4176 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
4178 btrfs_abort_transaction(trans
, root
, ret
);
4182 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
4183 inode_inc_iversion(dir
);
4184 dir
->i_mtime
= dir
->i_ctime
= current_fs_time(dir
->i_sb
);
4185 ret
= btrfs_update_inode_fallback(trans
, root
, dir
);
4187 btrfs_abort_transaction(trans
, root
, ret
);
4189 btrfs_free_path(path
);
4193 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
4195 struct inode
*inode
= d_inode(dentry
);
4197 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4198 struct btrfs_trans_handle
*trans
;
4200 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
4202 if (btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
)
4205 trans
= __unlink_start_trans(dir
);
4207 return PTR_ERR(trans
);
4209 if (unlikely(btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
4210 err
= btrfs_unlink_subvol(trans
, root
, dir
,
4211 BTRFS_I(inode
)->location
.objectid
,
4212 dentry
->d_name
.name
,
4213 dentry
->d_name
.len
);
4217 err
= btrfs_orphan_add(trans
, inode
);
4221 /* now the directory is empty */
4222 err
= btrfs_unlink_inode(trans
, root
, dir
, d_inode(dentry
),
4223 dentry
->d_name
.name
, dentry
->d_name
.len
);
4225 btrfs_i_size_write(inode
, 0);
4227 btrfs_end_transaction(trans
, root
);
4228 btrfs_btree_balance_dirty(root
);
4233 static int truncate_space_check(struct btrfs_trans_handle
*trans
,
4234 struct btrfs_root
*root
,
4240 * This is only used to apply pressure to the enospc system, we don't
4241 * intend to use this reservation at all.
4243 bytes_deleted
= btrfs_csum_bytes_to_leaves(root
, bytes_deleted
);
4244 bytes_deleted
*= root
->nodesize
;
4245 ret
= btrfs_block_rsv_add(root
, &root
->fs_info
->trans_block_rsv
,
4246 bytes_deleted
, BTRFS_RESERVE_NO_FLUSH
);
4248 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
4251 trans
->bytes_reserved
+= bytes_deleted
;
4257 static int truncate_inline_extent(struct inode
*inode
,
4258 struct btrfs_path
*path
,
4259 struct btrfs_key
*found_key
,
4263 struct extent_buffer
*leaf
= path
->nodes
[0];
4264 int slot
= path
->slots
[0];
4265 struct btrfs_file_extent_item
*fi
;
4266 u32 size
= (u32
)(new_size
- found_key
->offset
);
4267 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4269 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
4271 if (btrfs_file_extent_compression(leaf
, fi
) != BTRFS_COMPRESS_NONE
) {
4272 loff_t offset
= new_size
;
4273 loff_t page_end
= ALIGN(offset
, PAGE_SIZE
);
4276 * Zero out the remaining of the last page of our inline extent,
4277 * instead of directly truncating our inline extent here - that
4278 * would be much more complex (decompressing all the data, then
4279 * compressing the truncated data, which might be bigger than
4280 * the size of the inline extent, resize the extent, etc).
4281 * We release the path because to get the page we might need to
4282 * read the extent item from disk (data not in the page cache).
4284 btrfs_release_path(path
);
4285 return btrfs_truncate_block(inode
, offset
, page_end
- offset
,
4289 btrfs_set_file_extent_ram_bytes(leaf
, fi
, size
);
4290 size
= btrfs_file_extent_calc_inline_size(size
);
4291 btrfs_truncate_item(root
, path
, size
, 1);
4293 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4294 inode_sub_bytes(inode
, item_end
+ 1 - new_size
);
4300 * this can truncate away extent items, csum items and directory items.
4301 * It starts at a high offset and removes keys until it can't find
4302 * any higher than new_size
4304 * csum items that cross the new i_size are truncated to the new size
4307 * min_type is the minimum key type to truncate down to. If set to 0, this
4308 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4310 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
4311 struct btrfs_root
*root
,
4312 struct inode
*inode
,
4313 u64 new_size
, u32 min_type
)
4315 struct btrfs_path
*path
;
4316 struct extent_buffer
*leaf
;
4317 struct btrfs_file_extent_item
*fi
;
4318 struct btrfs_key key
;
4319 struct btrfs_key found_key
;
4320 u64 extent_start
= 0;
4321 u64 extent_num_bytes
= 0;
4322 u64 extent_offset
= 0;
4324 u64 last_size
= new_size
;
4325 u32 found_type
= (u8
)-1;
4328 int pending_del_nr
= 0;
4329 int pending_del_slot
= 0;
4330 int extent_type
= -1;
4333 u64 ino
= btrfs_ino(inode
);
4334 u64 bytes_deleted
= 0;
4336 bool should_throttle
= 0;
4337 bool should_end
= 0;
4339 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
4342 * for non-free space inodes and ref cows, we want to back off from
4345 if (!btrfs_is_free_space_inode(inode
) &&
4346 test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4349 path
= btrfs_alloc_path();
4352 path
->reada
= READA_BACK
;
4355 * We want to drop from the next block forward in case this new size is
4356 * not block aligned since we will be keeping the last block of the
4357 * extent just the way it is.
4359 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4360 root
== root
->fs_info
->tree_root
)
4361 btrfs_drop_extent_cache(inode
, ALIGN(new_size
,
4362 root
->sectorsize
), (u64
)-1, 0);
4365 * This function is also used to drop the items in the log tree before
4366 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4367 * it is used to drop the loged items. So we shouldn't kill the delayed
4370 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
4371 btrfs_kill_delayed_inode_items(inode
);
4374 key
.offset
= (u64
)-1;
4379 * with a 16K leaf size and 128MB extents, you can actually queue
4380 * up a huge file in a single leaf. Most of the time that
4381 * bytes_deleted is > 0, it will be huge by the time we get here
4383 if (be_nice
&& bytes_deleted
> SZ_32M
) {
4384 if (btrfs_should_end_transaction(trans
, root
)) {
4391 path
->leave_spinning
= 1;
4392 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
4399 /* there are no items in the tree for us to truncate, we're
4402 if (path
->slots
[0] == 0)
4409 leaf
= path
->nodes
[0];
4410 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4411 found_type
= found_key
.type
;
4413 if (found_key
.objectid
!= ino
)
4416 if (found_type
< min_type
)
4419 item_end
= found_key
.offset
;
4420 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
4421 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4422 struct btrfs_file_extent_item
);
4423 extent_type
= btrfs_file_extent_type(leaf
, fi
);
4424 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4426 btrfs_file_extent_num_bytes(leaf
, fi
);
4427 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4428 item_end
+= btrfs_file_extent_inline_len(leaf
,
4429 path
->slots
[0], fi
);
4433 if (found_type
> min_type
) {
4436 if (item_end
< new_size
)
4438 if (found_key
.offset
>= new_size
)
4444 /* FIXME, shrink the extent if the ref count is only 1 */
4445 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
4449 last_size
= found_key
.offset
;
4451 last_size
= new_size
;
4453 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4455 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4457 u64 orig_num_bytes
=
4458 btrfs_file_extent_num_bytes(leaf
, fi
);
4459 extent_num_bytes
= ALIGN(new_size
-
4462 btrfs_set_file_extent_num_bytes(leaf
, fi
,
4464 num_dec
= (orig_num_bytes
-
4466 if (test_bit(BTRFS_ROOT_REF_COWS
,
4469 inode_sub_bytes(inode
, num_dec
);
4470 btrfs_mark_buffer_dirty(leaf
);
4473 btrfs_file_extent_disk_num_bytes(leaf
,
4475 extent_offset
= found_key
.offset
-
4476 btrfs_file_extent_offset(leaf
, fi
);
4478 /* FIXME blocksize != 4096 */
4479 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
4480 if (extent_start
!= 0) {
4482 if (test_bit(BTRFS_ROOT_REF_COWS
,
4484 inode_sub_bytes(inode
, num_dec
);
4487 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4489 * we can't truncate inline items that have had
4493 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
4494 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
4497 * Need to release path in order to truncate a
4498 * compressed extent. So delete any accumulated
4499 * extent items so far.
4501 if (btrfs_file_extent_compression(leaf
, fi
) !=
4502 BTRFS_COMPRESS_NONE
&& pending_del_nr
) {
4503 err
= btrfs_del_items(trans
, root
, path
,
4507 btrfs_abort_transaction(trans
,
4515 err
= truncate_inline_extent(inode
, path
,
4520 btrfs_abort_transaction(trans
,
4524 } else if (test_bit(BTRFS_ROOT_REF_COWS
,
4526 inode_sub_bytes(inode
, item_end
+ 1 - new_size
);
4531 if (!pending_del_nr
) {
4532 /* no pending yet, add ourselves */
4533 pending_del_slot
= path
->slots
[0];
4535 } else if (pending_del_nr
&&
4536 path
->slots
[0] + 1 == pending_del_slot
) {
4537 /* hop on the pending chunk */
4539 pending_del_slot
= path
->slots
[0];
4546 should_throttle
= 0;
4549 (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4550 root
== root
->fs_info
->tree_root
)) {
4551 btrfs_set_path_blocking(path
);
4552 bytes_deleted
+= extent_num_bytes
;
4553 ret
= btrfs_free_extent(trans
, root
, extent_start
,
4554 extent_num_bytes
, 0,
4555 btrfs_header_owner(leaf
),
4556 ino
, extent_offset
);
4558 if (btrfs_should_throttle_delayed_refs(trans
, root
))
4559 btrfs_async_run_delayed_refs(root
,
4561 trans
->delayed_ref_updates
* 2, 0);
4563 if (truncate_space_check(trans
, root
,
4564 extent_num_bytes
)) {
4567 if (btrfs_should_throttle_delayed_refs(trans
,
4569 should_throttle
= 1;
4574 if (found_type
== BTRFS_INODE_ITEM_KEY
)
4577 if (path
->slots
[0] == 0 ||
4578 path
->slots
[0] != pending_del_slot
||
4579 should_throttle
|| should_end
) {
4580 if (pending_del_nr
) {
4581 ret
= btrfs_del_items(trans
, root
, path
,
4585 btrfs_abort_transaction(trans
,
4591 btrfs_release_path(path
);
4592 if (should_throttle
) {
4593 unsigned long updates
= trans
->delayed_ref_updates
;
4595 trans
->delayed_ref_updates
= 0;
4596 ret
= btrfs_run_delayed_refs(trans
, root
, updates
* 2);
4602 * if we failed to refill our space rsv, bail out
4603 * and let the transaction restart
4615 if (pending_del_nr
) {
4616 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
4619 btrfs_abort_transaction(trans
, root
, ret
);
4622 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
4623 btrfs_ordered_update_i_size(inode
, last_size
, NULL
);
4625 btrfs_free_path(path
);
4627 if (be_nice
&& bytes_deleted
> SZ_32M
) {
4628 unsigned long updates
= trans
->delayed_ref_updates
;
4630 trans
->delayed_ref_updates
= 0;
4631 ret
= btrfs_run_delayed_refs(trans
, root
, updates
* 2);
4640 * btrfs_truncate_block - read, zero a chunk and write a block
4641 * @inode - inode that we're zeroing
4642 * @from - the offset to start zeroing
4643 * @len - the length to zero, 0 to zero the entire range respective to the
4645 * @front - zero up to the offset instead of from the offset on
4647 * This will find the block for the "from" offset and cow the block and zero the
4648 * part we want to zero. This is used with truncate and hole punching.
4650 int btrfs_truncate_block(struct inode
*inode
, loff_t from
, loff_t len
,
4653 struct address_space
*mapping
= inode
->i_mapping
;
4654 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4655 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4656 struct btrfs_ordered_extent
*ordered
;
4657 struct extent_state
*cached_state
= NULL
;
4659 u32 blocksize
= root
->sectorsize
;
4660 pgoff_t index
= from
>> PAGE_SHIFT
;
4661 unsigned offset
= from
& (blocksize
- 1);
4663 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
4668 if ((offset
& (blocksize
- 1)) == 0 &&
4669 (!len
|| ((len
& (blocksize
- 1)) == 0)))
4672 ret
= btrfs_delalloc_reserve_space(inode
,
4673 round_down(from
, blocksize
), blocksize
);
4678 page
= find_or_create_page(mapping
, index
, mask
);
4680 btrfs_delalloc_release_space(inode
,
4681 round_down(from
, blocksize
),
4687 block_start
= round_down(from
, blocksize
);
4688 block_end
= block_start
+ blocksize
- 1;
4690 if (!PageUptodate(page
)) {
4691 ret
= btrfs_readpage(NULL
, page
);
4693 if (page
->mapping
!= mapping
) {
4698 if (!PageUptodate(page
)) {
4703 wait_on_page_writeback(page
);
4705 lock_extent_bits(io_tree
, block_start
, block_end
, &cached_state
);
4706 set_page_extent_mapped(page
);
4708 ordered
= btrfs_lookup_ordered_extent(inode
, block_start
);
4710 unlock_extent_cached(io_tree
, block_start
, block_end
,
4711 &cached_state
, GFP_NOFS
);
4714 btrfs_start_ordered_extent(inode
, ordered
, 1);
4715 btrfs_put_ordered_extent(ordered
);
4719 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, block_start
, block_end
,
4720 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4721 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
4722 0, 0, &cached_state
, GFP_NOFS
);
4724 ret
= btrfs_set_extent_delalloc(inode
, block_start
, block_end
,
4727 unlock_extent_cached(io_tree
, block_start
, block_end
,
4728 &cached_state
, GFP_NOFS
);
4732 if (offset
!= blocksize
) {
4734 len
= blocksize
- offset
;
4737 memset(kaddr
+ (block_start
- page_offset(page
)),
4740 memset(kaddr
+ (block_start
- page_offset(page
)) + offset
,
4742 flush_dcache_page(page
);
4745 ClearPageChecked(page
);
4746 set_page_dirty(page
);
4747 unlock_extent_cached(io_tree
, block_start
, block_end
, &cached_state
,
4752 btrfs_delalloc_release_space(inode
, block_start
,
4760 static int maybe_insert_hole(struct btrfs_root
*root
, struct inode
*inode
,
4761 u64 offset
, u64 len
)
4763 struct btrfs_trans_handle
*trans
;
4767 * Still need to make sure the inode looks like it's been updated so
4768 * that any holes get logged if we fsync.
4770 if (btrfs_fs_incompat(root
->fs_info
, NO_HOLES
)) {
4771 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
4772 BTRFS_I(inode
)->last_sub_trans
= root
->log_transid
;
4773 BTRFS_I(inode
)->last_log_commit
= root
->last_log_commit
;
4778 * 1 - for the one we're dropping
4779 * 1 - for the one we're adding
4780 * 1 - for updating the inode.
4782 trans
= btrfs_start_transaction(root
, 3);
4784 return PTR_ERR(trans
);
4786 ret
= btrfs_drop_extents(trans
, root
, inode
, offset
, offset
+ len
, 1);
4788 btrfs_abort_transaction(trans
, root
, ret
);
4789 btrfs_end_transaction(trans
, root
);
4793 ret
= btrfs_insert_file_extent(trans
, root
, btrfs_ino(inode
), offset
,
4794 0, 0, len
, 0, len
, 0, 0, 0);
4796 btrfs_abort_transaction(trans
, root
, ret
);
4798 btrfs_update_inode(trans
, root
, inode
);
4799 btrfs_end_transaction(trans
, root
);
4804 * This function puts in dummy file extents for the area we're creating a hole
4805 * for. So if we are truncating this file to a larger size we need to insert
4806 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4807 * the range between oldsize and size
4809 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
4811 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4812 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4813 struct extent_map
*em
= NULL
;
4814 struct extent_state
*cached_state
= NULL
;
4815 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4816 u64 hole_start
= ALIGN(oldsize
, root
->sectorsize
);
4817 u64 block_end
= ALIGN(size
, root
->sectorsize
);
4824 * If our size started in the middle of a block we need to zero out the
4825 * rest of the block before we expand the i_size, otherwise we could
4826 * expose stale data.
4828 err
= btrfs_truncate_block(inode
, oldsize
, 0, 0);
4832 if (size
<= hole_start
)
4836 struct btrfs_ordered_extent
*ordered
;
4838 lock_extent_bits(io_tree
, hole_start
, block_end
- 1,
4840 ordered
= btrfs_lookup_ordered_range(inode
, hole_start
,
4841 block_end
- hole_start
);
4844 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
4845 &cached_state
, GFP_NOFS
);
4846 btrfs_start_ordered_extent(inode
, ordered
, 1);
4847 btrfs_put_ordered_extent(ordered
);
4850 cur_offset
= hole_start
;
4852 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
4853 block_end
- cur_offset
, 0);
4859 last_byte
= min(extent_map_end(em
), block_end
);
4860 last_byte
= ALIGN(last_byte
, root
->sectorsize
);
4861 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
4862 struct extent_map
*hole_em
;
4863 hole_size
= last_byte
- cur_offset
;
4865 err
= maybe_insert_hole(root
, inode
, cur_offset
,
4869 btrfs_drop_extent_cache(inode
, cur_offset
,
4870 cur_offset
+ hole_size
- 1, 0);
4871 hole_em
= alloc_extent_map();
4873 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
4874 &BTRFS_I(inode
)->runtime_flags
);
4877 hole_em
->start
= cur_offset
;
4878 hole_em
->len
= hole_size
;
4879 hole_em
->orig_start
= cur_offset
;
4881 hole_em
->block_start
= EXTENT_MAP_HOLE
;
4882 hole_em
->block_len
= 0;
4883 hole_em
->orig_block_len
= 0;
4884 hole_em
->ram_bytes
= hole_size
;
4885 hole_em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4886 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
4887 hole_em
->generation
= root
->fs_info
->generation
;
4890 write_lock(&em_tree
->lock
);
4891 err
= add_extent_mapping(em_tree
, hole_em
, 1);
4892 write_unlock(&em_tree
->lock
);
4895 btrfs_drop_extent_cache(inode
, cur_offset
,
4899 free_extent_map(hole_em
);
4902 free_extent_map(em
);
4904 cur_offset
= last_byte
;
4905 if (cur_offset
>= block_end
)
4908 free_extent_map(em
);
4909 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
4914 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
)
4916 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4917 struct btrfs_trans_handle
*trans
;
4918 loff_t oldsize
= i_size_read(inode
);
4919 loff_t newsize
= attr
->ia_size
;
4920 int mask
= attr
->ia_valid
;
4924 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4925 * special case where we need to update the times despite not having
4926 * these flags set. For all other operations the VFS set these flags
4927 * explicitly if it wants a timestamp update.
4929 if (newsize
!= oldsize
) {
4930 inode_inc_iversion(inode
);
4931 if (!(mask
& (ATTR_CTIME
| ATTR_MTIME
)))
4932 inode
->i_ctime
= inode
->i_mtime
=
4933 current_fs_time(inode
->i_sb
);
4936 if (newsize
> oldsize
) {
4938 * Don't do an expanding truncate while snapshoting is ongoing.
4939 * This is to ensure the snapshot captures a fully consistent
4940 * state of this file - if the snapshot captures this expanding
4941 * truncation, it must capture all writes that happened before
4944 btrfs_wait_for_snapshot_creation(root
);
4945 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
4947 btrfs_end_write_no_snapshoting(root
);
4951 trans
= btrfs_start_transaction(root
, 1);
4952 if (IS_ERR(trans
)) {
4953 btrfs_end_write_no_snapshoting(root
);
4954 return PTR_ERR(trans
);
4957 i_size_write(inode
, newsize
);
4958 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
4959 pagecache_isize_extended(inode
, oldsize
, newsize
);
4960 ret
= btrfs_update_inode(trans
, root
, inode
);
4961 btrfs_end_write_no_snapshoting(root
);
4962 btrfs_end_transaction(trans
, root
);
4966 * We're truncating a file that used to have good data down to
4967 * zero. Make sure it gets into the ordered flush list so that
4968 * any new writes get down to disk quickly.
4971 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
4972 &BTRFS_I(inode
)->runtime_flags
);
4975 * 1 for the orphan item we're going to add
4976 * 1 for the orphan item deletion.
4978 trans
= btrfs_start_transaction(root
, 2);
4980 return PTR_ERR(trans
);
4983 * We need to do this in case we fail at _any_ point during the
4984 * actual truncate. Once we do the truncate_setsize we could
4985 * invalidate pages which forces any outstanding ordered io to
4986 * be instantly completed which will give us extents that need
4987 * to be truncated. If we fail to get an orphan inode down we
4988 * could have left over extents that were never meant to live,
4989 * so we need to guarantee from this point on that everything
4990 * will be consistent.
4992 ret
= btrfs_orphan_add(trans
, inode
);
4993 btrfs_end_transaction(trans
, root
);
4997 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4998 truncate_setsize(inode
, newsize
);
5000 /* Disable nonlocked read DIO to avoid the end less truncate */
5001 btrfs_inode_block_unlocked_dio(inode
);
5002 inode_dio_wait(inode
);
5003 btrfs_inode_resume_unlocked_dio(inode
);
5005 ret
= btrfs_truncate(inode
);
5006 if (ret
&& inode
->i_nlink
) {
5010 * failed to truncate, disk_i_size is only adjusted down
5011 * as we remove extents, so it should represent the true
5012 * size of the inode, so reset the in memory size and
5013 * delete our orphan entry.
5015 trans
= btrfs_join_transaction(root
);
5016 if (IS_ERR(trans
)) {
5017 btrfs_orphan_del(NULL
, inode
);
5020 i_size_write(inode
, BTRFS_I(inode
)->disk_i_size
);
5021 err
= btrfs_orphan_del(trans
, inode
);
5023 btrfs_abort_transaction(trans
, root
, err
);
5024 btrfs_end_transaction(trans
, root
);
5031 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5033 struct inode
*inode
= d_inode(dentry
);
5034 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5037 if (btrfs_root_readonly(root
))
5040 err
= inode_change_ok(inode
, attr
);
5044 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
5045 err
= btrfs_setsize(inode
, attr
);
5050 if (attr
->ia_valid
) {
5051 setattr_copy(inode
, attr
);
5052 inode_inc_iversion(inode
);
5053 err
= btrfs_dirty_inode(inode
);
5055 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
5056 err
= posix_acl_chmod(inode
, inode
->i_mode
);
5063 * While truncating the inode pages during eviction, we get the VFS calling
5064 * btrfs_invalidatepage() against each page of the inode. This is slow because
5065 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5066 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5067 * extent_state structures over and over, wasting lots of time.
5069 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5070 * those expensive operations on a per page basis and do only the ordered io
5071 * finishing, while we release here the extent_map and extent_state structures,
5072 * without the excessive merging and splitting.
5074 static void evict_inode_truncate_pages(struct inode
*inode
)
5076 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5077 struct extent_map_tree
*map_tree
= &BTRFS_I(inode
)->extent_tree
;
5078 struct rb_node
*node
;
5080 ASSERT(inode
->i_state
& I_FREEING
);
5081 truncate_inode_pages_final(&inode
->i_data
);
5083 write_lock(&map_tree
->lock
);
5084 while (!RB_EMPTY_ROOT(&map_tree
->map
)) {
5085 struct extent_map
*em
;
5087 node
= rb_first(&map_tree
->map
);
5088 em
= rb_entry(node
, struct extent_map
, rb_node
);
5089 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5090 clear_bit(EXTENT_FLAG_LOGGING
, &em
->flags
);
5091 remove_extent_mapping(map_tree
, em
);
5092 free_extent_map(em
);
5093 if (need_resched()) {
5094 write_unlock(&map_tree
->lock
);
5096 write_lock(&map_tree
->lock
);
5099 write_unlock(&map_tree
->lock
);
5102 * Keep looping until we have no more ranges in the io tree.
5103 * We can have ongoing bios started by readpages (called from readahead)
5104 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5105 * still in progress (unlocked the pages in the bio but did not yet
5106 * unlocked the ranges in the io tree). Therefore this means some
5107 * ranges can still be locked and eviction started because before
5108 * submitting those bios, which are executed by a separate task (work
5109 * queue kthread), inode references (inode->i_count) were not taken
5110 * (which would be dropped in the end io callback of each bio).
5111 * Therefore here we effectively end up waiting for those bios and
5112 * anyone else holding locked ranges without having bumped the inode's
5113 * reference count - if we don't do it, when they access the inode's
5114 * io_tree to unlock a range it may be too late, leading to an
5115 * use-after-free issue.
5117 spin_lock(&io_tree
->lock
);
5118 while (!RB_EMPTY_ROOT(&io_tree
->state
)) {
5119 struct extent_state
*state
;
5120 struct extent_state
*cached_state
= NULL
;
5124 node
= rb_first(&io_tree
->state
);
5125 state
= rb_entry(node
, struct extent_state
, rb_node
);
5126 start
= state
->start
;
5128 spin_unlock(&io_tree
->lock
);
5130 lock_extent_bits(io_tree
, start
, end
, &cached_state
);
5133 * If still has DELALLOC flag, the extent didn't reach disk,
5134 * and its reserved space won't be freed by delayed_ref.
5135 * So we need to free its reserved space here.
5136 * (Refer to comment in btrfs_invalidatepage, case 2)
5138 * Note, end is the bytenr of last byte, so we need + 1 here.
5140 if (state
->state
& EXTENT_DELALLOC
)
5141 btrfs_qgroup_free_data(inode
, start
, end
- start
+ 1);
5143 clear_extent_bit(io_tree
, start
, end
,
5144 EXTENT_LOCKED
| EXTENT_DIRTY
|
5145 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
5146 EXTENT_DEFRAG
, 1, 1,
5147 &cached_state
, GFP_NOFS
);
5150 spin_lock(&io_tree
->lock
);
5152 spin_unlock(&io_tree
->lock
);
5155 void btrfs_evict_inode(struct inode
*inode
)
5157 struct btrfs_trans_handle
*trans
;
5158 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5159 struct btrfs_block_rsv
*rsv
, *global_rsv
;
5160 int steal_from_global
= 0;
5161 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
5164 trace_btrfs_inode_evict(inode
);
5166 evict_inode_truncate_pages(inode
);
5168 if (inode
->i_nlink
&&
5169 ((btrfs_root_refs(&root
->root_item
) != 0 &&
5170 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
) ||
5171 btrfs_is_free_space_inode(inode
)))
5174 if (is_bad_inode(inode
)) {
5175 btrfs_orphan_del(NULL
, inode
);
5178 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5179 if (!special_file(inode
->i_mode
))
5180 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
5182 btrfs_free_io_failure_record(inode
, 0, (u64
)-1);
5184 if (root
->fs_info
->log_root_recovering
) {
5185 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
5186 &BTRFS_I(inode
)->runtime_flags
));
5190 if (inode
->i_nlink
> 0) {
5191 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0 &&
5192 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
);
5196 ret
= btrfs_commit_inode_delayed_inode(inode
);
5198 btrfs_orphan_del(NULL
, inode
);
5202 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
5204 btrfs_orphan_del(NULL
, inode
);
5207 rsv
->size
= min_size
;
5209 global_rsv
= &root
->fs_info
->global_block_rsv
;
5211 btrfs_i_size_write(inode
, 0);
5214 * This is a bit simpler than btrfs_truncate since we've already
5215 * reserved our space for our orphan item in the unlink, so we just
5216 * need to reserve some slack space in case we add bytes and update
5217 * inode item when doing the truncate.
5220 ret
= btrfs_block_rsv_refill(root
, rsv
, min_size
,
5221 BTRFS_RESERVE_FLUSH_LIMIT
);
5224 * Try and steal from the global reserve since we will
5225 * likely not use this space anyway, we want to try as
5226 * hard as possible to get this to work.
5229 steal_from_global
++;
5231 steal_from_global
= 0;
5235 * steal_from_global == 0: we reserved stuff, hooray!
5236 * steal_from_global == 1: we didn't reserve stuff, boo!
5237 * steal_from_global == 2: we've committed, still not a lot of
5238 * room but maybe we'll have room in the global reserve this
5240 * steal_from_global == 3: abandon all hope!
5242 if (steal_from_global
> 2) {
5243 btrfs_warn(root
->fs_info
,
5244 "Could not get space for a delete, will truncate on mount %d",
5246 btrfs_orphan_del(NULL
, inode
);
5247 btrfs_free_block_rsv(root
, rsv
);
5251 trans
= btrfs_join_transaction(root
);
5252 if (IS_ERR(trans
)) {
5253 btrfs_orphan_del(NULL
, inode
);
5254 btrfs_free_block_rsv(root
, rsv
);
5259 * We can't just steal from the global reserve, we need to make
5260 * sure there is room to do it, if not we need to commit and try
5263 if (steal_from_global
) {
5264 if (!btrfs_check_space_for_delayed_refs(trans
, root
))
5265 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
,
5272 * Couldn't steal from the global reserve, we have too much
5273 * pending stuff built up, commit the transaction and try it
5277 ret
= btrfs_commit_transaction(trans
, root
);
5279 btrfs_orphan_del(NULL
, inode
);
5280 btrfs_free_block_rsv(root
, rsv
);
5285 steal_from_global
= 0;
5288 trans
->block_rsv
= rsv
;
5290 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
5291 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
)
5294 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
5295 btrfs_end_transaction(trans
, root
);
5297 btrfs_btree_balance_dirty(root
);
5300 btrfs_free_block_rsv(root
, rsv
);
5303 * Errors here aren't a big deal, it just means we leave orphan items
5304 * in the tree. They will be cleaned up on the next mount.
5307 trans
->block_rsv
= root
->orphan_block_rsv
;
5308 btrfs_orphan_del(trans
, inode
);
5310 btrfs_orphan_del(NULL
, inode
);
5313 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
5314 if (!(root
== root
->fs_info
->tree_root
||
5315 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
5316 btrfs_return_ino(root
, btrfs_ino(inode
));
5318 btrfs_end_transaction(trans
, root
);
5319 btrfs_btree_balance_dirty(root
);
5321 btrfs_remove_delayed_node(inode
);
5326 * this returns the key found in the dir entry in the location pointer.
5327 * If no dir entries were found, location->objectid is 0.
5329 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
5330 struct btrfs_key
*location
)
5332 const char *name
= dentry
->d_name
.name
;
5333 int namelen
= dentry
->d_name
.len
;
5334 struct btrfs_dir_item
*di
;
5335 struct btrfs_path
*path
;
5336 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5339 path
= btrfs_alloc_path();
5343 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
), name
,
5348 if (IS_ERR_OR_NULL(di
))
5351 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
5353 btrfs_free_path(path
);
5356 location
->objectid
= 0;
5361 * when we hit a tree root in a directory, the btrfs part of the inode
5362 * needs to be changed to reflect the root directory of the tree root. This
5363 * is kind of like crossing a mount point.
5365 static int fixup_tree_root_location(struct btrfs_root
*root
,
5367 struct dentry
*dentry
,
5368 struct btrfs_key
*location
,
5369 struct btrfs_root
**sub_root
)
5371 struct btrfs_path
*path
;
5372 struct btrfs_root
*new_root
;
5373 struct btrfs_root_ref
*ref
;
5374 struct extent_buffer
*leaf
;
5375 struct btrfs_key key
;
5379 path
= btrfs_alloc_path();
5386 key
.objectid
= BTRFS_I(dir
)->root
->root_key
.objectid
;
5387 key
.type
= BTRFS_ROOT_REF_KEY
;
5388 key
.offset
= location
->objectid
;
5390 ret
= btrfs_search_slot(NULL
, root
->fs_info
->tree_root
, &key
, path
,
5398 leaf
= path
->nodes
[0];
5399 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
5400 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
5401 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
5404 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
5405 (unsigned long)(ref
+ 1),
5406 dentry
->d_name
.len
);
5410 btrfs_release_path(path
);
5412 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
5413 if (IS_ERR(new_root
)) {
5414 err
= PTR_ERR(new_root
);
5418 *sub_root
= new_root
;
5419 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
5420 location
->type
= BTRFS_INODE_ITEM_KEY
;
5421 location
->offset
= 0;
5424 btrfs_free_path(path
);
5428 static void inode_tree_add(struct inode
*inode
)
5430 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5431 struct btrfs_inode
*entry
;
5433 struct rb_node
*parent
;
5434 struct rb_node
*new = &BTRFS_I(inode
)->rb_node
;
5435 u64 ino
= btrfs_ino(inode
);
5437 if (inode_unhashed(inode
))
5440 spin_lock(&root
->inode_lock
);
5441 p
= &root
->inode_tree
.rb_node
;
5444 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
5446 if (ino
< btrfs_ino(&entry
->vfs_inode
))
5447 p
= &parent
->rb_left
;
5448 else if (ino
> btrfs_ino(&entry
->vfs_inode
))
5449 p
= &parent
->rb_right
;
5451 WARN_ON(!(entry
->vfs_inode
.i_state
&
5452 (I_WILL_FREE
| I_FREEING
)));
5453 rb_replace_node(parent
, new, &root
->inode_tree
);
5454 RB_CLEAR_NODE(parent
);
5455 spin_unlock(&root
->inode_lock
);
5459 rb_link_node(new, parent
, p
);
5460 rb_insert_color(new, &root
->inode_tree
);
5461 spin_unlock(&root
->inode_lock
);
5464 static void inode_tree_del(struct inode
*inode
)
5466 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5469 spin_lock(&root
->inode_lock
);
5470 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
5471 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
5472 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
5473 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5475 spin_unlock(&root
->inode_lock
);
5477 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0) {
5478 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
5479 spin_lock(&root
->inode_lock
);
5480 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5481 spin_unlock(&root
->inode_lock
);
5483 btrfs_add_dead_root(root
);
5487 void btrfs_invalidate_inodes(struct btrfs_root
*root
)
5489 struct rb_node
*node
;
5490 struct rb_node
*prev
;
5491 struct btrfs_inode
*entry
;
5492 struct inode
*inode
;
5495 if (!test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
5496 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
5498 spin_lock(&root
->inode_lock
);
5500 node
= root
->inode_tree
.rb_node
;
5504 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5506 if (objectid
< btrfs_ino(&entry
->vfs_inode
))
5507 node
= node
->rb_left
;
5508 else if (objectid
> btrfs_ino(&entry
->vfs_inode
))
5509 node
= node
->rb_right
;
5515 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
5516 if (objectid
<= btrfs_ino(&entry
->vfs_inode
)) {
5520 prev
= rb_next(prev
);
5524 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5525 objectid
= btrfs_ino(&entry
->vfs_inode
) + 1;
5526 inode
= igrab(&entry
->vfs_inode
);
5528 spin_unlock(&root
->inode_lock
);
5529 if (atomic_read(&inode
->i_count
) > 1)
5530 d_prune_aliases(inode
);
5532 * btrfs_drop_inode will have it removed from
5533 * the inode cache when its usage count
5538 spin_lock(&root
->inode_lock
);
5542 if (cond_resched_lock(&root
->inode_lock
))
5545 node
= rb_next(node
);
5547 spin_unlock(&root
->inode_lock
);
5550 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
5552 struct btrfs_iget_args
*args
= p
;
5553 inode
->i_ino
= args
->location
->objectid
;
5554 memcpy(&BTRFS_I(inode
)->location
, args
->location
,
5555 sizeof(*args
->location
));
5556 BTRFS_I(inode
)->root
= args
->root
;
5560 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
5562 struct btrfs_iget_args
*args
= opaque
;
5563 return args
->location
->objectid
== BTRFS_I(inode
)->location
.objectid
&&
5564 args
->root
== BTRFS_I(inode
)->root
;
5567 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
5568 struct btrfs_key
*location
,
5569 struct btrfs_root
*root
)
5571 struct inode
*inode
;
5572 struct btrfs_iget_args args
;
5573 unsigned long hashval
= btrfs_inode_hash(location
->objectid
, root
);
5575 args
.location
= location
;
5578 inode
= iget5_locked(s
, hashval
, btrfs_find_actor
,
5579 btrfs_init_locked_inode
,
5584 /* Get an inode object given its location and corresponding root.
5585 * Returns in *is_new if the inode was read from disk
5587 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
5588 struct btrfs_root
*root
, int *new)
5590 struct inode
*inode
;
5592 inode
= btrfs_iget_locked(s
, location
, root
);
5594 return ERR_PTR(-ENOMEM
);
5596 if (inode
->i_state
& I_NEW
) {
5597 btrfs_read_locked_inode(inode
);
5598 if (!is_bad_inode(inode
)) {
5599 inode_tree_add(inode
);
5600 unlock_new_inode(inode
);
5604 unlock_new_inode(inode
);
5606 inode
= ERR_PTR(-ESTALE
);
5613 static struct inode
*new_simple_dir(struct super_block
*s
,
5614 struct btrfs_key
*key
,
5615 struct btrfs_root
*root
)
5617 struct inode
*inode
= new_inode(s
);
5620 return ERR_PTR(-ENOMEM
);
5622 BTRFS_I(inode
)->root
= root
;
5623 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
5624 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
5626 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
5627 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
5628 inode
->i_fop
= &simple_dir_operations
;
5629 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
5630 inode
->i_mtime
= current_fs_time(inode
->i_sb
);
5631 inode
->i_atime
= inode
->i_mtime
;
5632 inode
->i_ctime
= inode
->i_mtime
;
5633 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
5638 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
5640 struct inode
*inode
;
5641 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5642 struct btrfs_root
*sub_root
= root
;
5643 struct btrfs_key location
;
5647 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
5648 return ERR_PTR(-ENAMETOOLONG
);
5650 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
5652 return ERR_PTR(ret
);
5654 if (location
.objectid
== 0)
5655 return ERR_PTR(-ENOENT
);
5657 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
5658 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
5662 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
5664 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
5665 ret
= fixup_tree_root_location(root
, dir
, dentry
,
5666 &location
, &sub_root
);
5669 inode
= ERR_PTR(ret
);
5671 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
5673 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
5675 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
5677 if (!IS_ERR(inode
) && root
!= sub_root
) {
5678 down_read(&root
->fs_info
->cleanup_work_sem
);
5679 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
5680 ret
= btrfs_orphan_cleanup(sub_root
);
5681 up_read(&root
->fs_info
->cleanup_work_sem
);
5684 inode
= ERR_PTR(ret
);
5691 static int btrfs_dentry_delete(const struct dentry
*dentry
)
5693 struct btrfs_root
*root
;
5694 struct inode
*inode
= d_inode(dentry
);
5696 if (!inode
&& !IS_ROOT(dentry
))
5697 inode
= d_inode(dentry
->d_parent
);
5700 root
= BTRFS_I(inode
)->root
;
5701 if (btrfs_root_refs(&root
->root_item
) == 0)
5704 if (btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
5710 static void btrfs_dentry_release(struct dentry
*dentry
)
5712 kfree(dentry
->d_fsdata
);
5715 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
5718 struct inode
*inode
;
5720 inode
= btrfs_lookup_dentry(dir
, dentry
);
5721 if (IS_ERR(inode
)) {
5722 if (PTR_ERR(inode
) == -ENOENT
)
5725 return ERR_CAST(inode
);
5728 return d_splice_alias(inode
, dentry
);
5731 unsigned char btrfs_filetype_table
[] = {
5732 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
5735 static int btrfs_real_readdir(struct file
*file
, struct dir_context
*ctx
)
5737 struct inode
*inode
= file_inode(file
);
5738 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5739 struct btrfs_item
*item
;
5740 struct btrfs_dir_item
*di
;
5741 struct btrfs_key key
;
5742 struct btrfs_key found_key
;
5743 struct btrfs_path
*path
;
5744 struct list_head ins_list
;
5745 struct list_head del_list
;
5747 struct extent_buffer
*leaf
;
5749 unsigned char d_type
;
5754 int key_type
= BTRFS_DIR_INDEX_KEY
;
5758 int is_curr
= 0; /* ctx->pos points to the current index? */
5762 /* FIXME, use a real flag for deciding about the key type */
5763 if (root
->fs_info
->tree_root
== root
)
5764 key_type
= BTRFS_DIR_ITEM_KEY
;
5766 if (!dir_emit_dots(file
, ctx
))
5769 path
= btrfs_alloc_path();
5773 path
->reada
= READA_FORWARD
;
5775 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5776 INIT_LIST_HEAD(&ins_list
);
5777 INIT_LIST_HEAD(&del_list
);
5778 put
= btrfs_readdir_get_delayed_items(inode
, &ins_list
,
5782 key
.type
= key_type
;
5783 key
.offset
= ctx
->pos
;
5784 key
.objectid
= btrfs_ino(inode
);
5786 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5792 leaf
= path
->nodes
[0];
5793 slot
= path
->slots
[0];
5794 if (slot
>= btrfs_header_nritems(leaf
)) {
5795 ret
= btrfs_next_leaf(root
, path
);
5803 item
= btrfs_item_nr(slot
);
5804 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5806 if (found_key
.objectid
!= key
.objectid
)
5808 if (found_key
.type
!= key_type
)
5810 if (found_key
.offset
< ctx
->pos
)
5812 if (key_type
== BTRFS_DIR_INDEX_KEY
&&
5813 btrfs_should_delete_dir_index(&del_list
,
5817 ctx
->pos
= found_key
.offset
;
5820 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
5822 di_total
= btrfs_item_size(leaf
, item
);
5824 while (di_cur
< di_total
) {
5825 struct btrfs_key location
;
5827 if (verify_dir_item(root
, leaf
, di
))
5830 name_len
= btrfs_dir_name_len(leaf
, di
);
5831 if (name_len
<= sizeof(tmp_name
)) {
5832 name_ptr
= tmp_name
;
5834 name_ptr
= kmalloc(name_len
, GFP_KERNEL
);
5840 read_extent_buffer(leaf
, name_ptr
,
5841 (unsigned long)(di
+ 1), name_len
);
5843 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
5844 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
5847 /* is this a reference to our own snapshot? If so
5850 * In contrast to old kernels, we insert the snapshot's
5851 * dir item and dir index after it has been created, so
5852 * we won't find a reference to our own snapshot. We
5853 * still keep the following code for backward
5856 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
5857 location
.objectid
== root
->root_key
.objectid
) {
5861 over
= !dir_emit(ctx
, name_ptr
, name_len
,
5862 location
.objectid
, d_type
);
5865 if (name_ptr
!= tmp_name
)
5871 di_len
= btrfs_dir_name_len(leaf
, di
) +
5872 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
5874 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
5880 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5883 ret
= btrfs_readdir_delayed_dir_index(ctx
, &ins_list
, &emitted
);
5889 * If we haven't emitted any dir entry, we must not touch ctx->pos as
5890 * it was was set to the termination value in previous call. We assume
5891 * that "." and ".." were emitted if we reach this point and set the
5892 * termination value as well for an empty directory.
5894 if (ctx
->pos
> 2 && !emitted
)
5897 /* Reached end of directory/root. Bump pos past the last item. */
5901 * Stop new entries from being returned after we return the last
5904 * New directory entries are assigned a strictly increasing
5905 * offset. This means that new entries created during readdir
5906 * are *guaranteed* to be seen in the future by that readdir.
5907 * This has broken buggy programs which operate on names as
5908 * they're returned by readdir. Until we re-use freed offsets
5909 * we have this hack to stop new entries from being returned
5910 * under the assumption that they'll never reach this huge
5913 * This is being careful not to overflow 32bit loff_t unless the
5914 * last entry requires it because doing so has broken 32bit apps
5917 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5918 if (ctx
->pos
>= INT_MAX
)
5919 ctx
->pos
= LLONG_MAX
;
5927 btrfs_readdir_put_delayed_items(inode
, &ins_list
, &del_list
);
5928 btrfs_free_path(path
);
5932 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5934 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5935 struct btrfs_trans_handle
*trans
;
5937 bool nolock
= false;
5939 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5942 if (btrfs_fs_closing(root
->fs_info
) && btrfs_is_free_space_inode(inode
))
5945 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
5947 trans
= btrfs_join_transaction_nolock(root
);
5949 trans
= btrfs_join_transaction(root
);
5951 return PTR_ERR(trans
);
5952 ret
= btrfs_commit_transaction(trans
, root
);
5958 * This is somewhat expensive, updating the tree every time the
5959 * inode changes. But, it is most likely to find the inode in cache.
5960 * FIXME, needs more benchmarking...there are no reasons other than performance
5961 * to keep or drop this code.
5963 static int btrfs_dirty_inode(struct inode
*inode
)
5965 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5966 struct btrfs_trans_handle
*trans
;
5969 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5972 trans
= btrfs_join_transaction(root
);
5974 return PTR_ERR(trans
);
5976 ret
= btrfs_update_inode(trans
, root
, inode
);
5977 if (ret
&& ret
== -ENOSPC
) {
5978 /* whoops, lets try again with the full transaction */
5979 btrfs_end_transaction(trans
, root
);
5980 trans
= btrfs_start_transaction(root
, 1);
5982 return PTR_ERR(trans
);
5984 ret
= btrfs_update_inode(trans
, root
, inode
);
5986 btrfs_end_transaction(trans
, root
);
5987 if (BTRFS_I(inode
)->delayed_node
)
5988 btrfs_balance_delayed_items(root
);
5994 * This is a copy of file_update_time. We need this so we can return error on
5995 * ENOSPC for updating the inode in the case of file write and mmap writes.
5997 static int btrfs_update_time(struct inode
*inode
, struct timespec
*now
,
6000 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6002 if (btrfs_root_readonly(root
))
6005 if (flags
& S_VERSION
)
6006 inode_inc_iversion(inode
);
6007 if (flags
& S_CTIME
)
6008 inode
->i_ctime
= *now
;
6009 if (flags
& S_MTIME
)
6010 inode
->i_mtime
= *now
;
6011 if (flags
& S_ATIME
)
6012 inode
->i_atime
= *now
;
6013 return btrfs_dirty_inode(inode
);
6017 * find the highest existing sequence number in a directory
6018 * and then set the in-memory index_cnt variable to reflect
6019 * free sequence numbers
6021 static int btrfs_set_inode_index_count(struct inode
*inode
)
6023 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6024 struct btrfs_key key
, found_key
;
6025 struct btrfs_path
*path
;
6026 struct extent_buffer
*leaf
;
6029 key
.objectid
= btrfs_ino(inode
);
6030 key
.type
= BTRFS_DIR_INDEX_KEY
;
6031 key
.offset
= (u64
)-1;
6033 path
= btrfs_alloc_path();
6037 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6040 /* FIXME: we should be able to handle this */
6046 * MAGIC NUMBER EXPLANATION:
6047 * since we search a directory based on f_pos we have to start at 2
6048 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6049 * else has to start at 2
6051 if (path
->slots
[0] == 0) {
6052 BTRFS_I(inode
)->index_cnt
= 2;
6058 leaf
= path
->nodes
[0];
6059 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6061 if (found_key
.objectid
!= btrfs_ino(inode
) ||
6062 found_key
.type
!= BTRFS_DIR_INDEX_KEY
) {
6063 BTRFS_I(inode
)->index_cnt
= 2;
6067 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
6069 btrfs_free_path(path
);
6074 * helper to find a free sequence number in a given directory. This current
6075 * code is very simple, later versions will do smarter things in the btree
6077 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
6081 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
6082 ret
= btrfs_inode_delayed_dir_index_count(dir
);
6084 ret
= btrfs_set_inode_index_count(dir
);
6090 *index
= BTRFS_I(dir
)->index_cnt
;
6091 BTRFS_I(dir
)->index_cnt
++;
6096 static int btrfs_insert_inode_locked(struct inode
*inode
)
6098 struct btrfs_iget_args args
;
6099 args
.location
= &BTRFS_I(inode
)->location
;
6100 args
.root
= BTRFS_I(inode
)->root
;
6102 return insert_inode_locked4(inode
,
6103 btrfs_inode_hash(inode
->i_ino
, BTRFS_I(inode
)->root
),
6104 btrfs_find_actor
, &args
);
6107 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
6108 struct btrfs_root
*root
,
6110 const char *name
, int name_len
,
6111 u64 ref_objectid
, u64 objectid
,
6112 umode_t mode
, u64
*index
)
6114 struct inode
*inode
;
6115 struct btrfs_inode_item
*inode_item
;
6116 struct btrfs_key
*location
;
6117 struct btrfs_path
*path
;
6118 struct btrfs_inode_ref
*ref
;
6119 struct btrfs_key key
[2];
6121 int nitems
= name
? 2 : 1;
6125 path
= btrfs_alloc_path();
6127 return ERR_PTR(-ENOMEM
);
6129 inode
= new_inode(root
->fs_info
->sb
);
6131 btrfs_free_path(path
);
6132 return ERR_PTR(-ENOMEM
);
6136 * O_TMPFILE, set link count to 0, so that after this point,
6137 * we fill in an inode item with the correct link count.
6140 set_nlink(inode
, 0);
6143 * we have to initialize this early, so we can reclaim the inode
6144 * number if we fail afterwards in this function.
6146 inode
->i_ino
= objectid
;
6149 trace_btrfs_inode_request(dir
);
6151 ret
= btrfs_set_inode_index(dir
, index
);
6153 btrfs_free_path(path
);
6155 return ERR_PTR(ret
);
6161 * index_cnt is ignored for everything but a dir,
6162 * btrfs_get_inode_index_count has an explanation for the magic
6165 BTRFS_I(inode
)->index_cnt
= 2;
6166 BTRFS_I(inode
)->dir_index
= *index
;
6167 BTRFS_I(inode
)->root
= root
;
6168 BTRFS_I(inode
)->generation
= trans
->transid
;
6169 inode
->i_generation
= BTRFS_I(inode
)->generation
;
6172 * We could have gotten an inode number from somebody who was fsynced
6173 * and then removed in this same transaction, so let's just set full
6174 * sync since it will be a full sync anyway and this will blow away the
6175 * old info in the log.
6177 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
6179 key
[0].objectid
= objectid
;
6180 key
[0].type
= BTRFS_INODE_ITEM_KEY
;
6183 sizes
[0] = sizeof(struct btrfs_inode_item
);
6187 * Start new inodes with an inode_ref. This is slightly more
6188 * efficient for small numbers of hard links since they will
6189 * be packed into one item. Extended refs will kick in if we
6190 * add more hard links than can fit in the ref item.
6192 key
[1].objectid
= objectid
;
6193 key
[1].type
= BTRFS_INODE_REF_KEY
;
6194 key
[1].offset
= ref_objectid
;
6196 sizes
[1] = name_len
+ sizeof(*ref
);
6199 location
= &BTRFS_I(inode
)->location
;
6200 location
->objectid
= objectid
;
6201 location
->offset
= 0;
6202 location
->type
= BTRFS_INODE_ITEM_KEY
;
6204 ret
= btrfs_insert_inode_locked(inode
);
6208 path
->leave_spinning
= 1;
6209 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, nitems
);
6213 inode_init_owner(inode
, dir
, mode
);
6214 inode_set_bytes(inode
, 0);
6216 inode
->i_mtime
= current_fs_time(inode
->i_sb
);
6217 inode
->i_atime
= inode
->i_mtime
;
6218 inode
->i_ctime
= inode
->i_mtime
;
6219 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
6221 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
6222 struct btrfs_inode_item
);
6223 memset_extent_buffer(path
->nodes
[0], 0, (unsigned long)inode_item
,
6224 sizeof(*inode_item
));
6225 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
6228 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
6229 struct btrfs_inode_ref
);
6230 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
6231 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
6232 ptr
= (unsigned long)(ref
+ 1);
6233 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
6236 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6237 btrfs_free_path(path
);
6239 btrfs_inherit_iflags(inode
, dir
);
6241 if (S_ISREG(mode
)) {
6242 if (btrfs_test_opt(root
, NODATASUM
))
6243 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
6244 if (btrfs_test_opt(root
, NODATACOW
))
6245 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
|
6246 BTRFS_INODE_NODATASUM
;
6249 inode_tree_add(inode
);
6251 trace_btrfs_inode_new(inode
);
6252 btrfs_set_inode_last_trans(trans
, inode
);
6254 btrfs_update_root_times(trans
, root
);
6256 ret
= btrfs_inode_inherit_props(trans
, inode
, dir
);
6258 btrfs_err(root
->fs_info
,
6259 "error inheriting props for ino %llu (root %llu): %d",
6260 btrfs_ino(inode
), root
->root_key
.objectid
, ret
);
6265 unlock_new_inode(inode
);
6268 BTRFS_I(dir
)->index_cnt
--;
6269 btrfs_free_path(path
);
6271 return ERR_PTR(ret
);
6274 static inline u8
btrfs_inode_type(struct inode
*inode
)
6276 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
6280 * utility function to add 'inode' into 'parent_inode' with
6281 * a give name and a given sequence number.
6282 * if 'add_backref' is true, also insert a backref from the
6283 * inode to the parent directory.
6285 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
6286 struct inode
*parent_inode
, struct inode
*inode
,
6287 const char *name
, int name_len
, int add_backref
, u64 index
)
6290 struct btrfs_key key
;
6291 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
6292 u64 ino
= btrfs_ino(inode
);
6293 u64 parent_ino
= btrfs_ino(parent_inode
);
6295 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6296 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
6299 key
.type
= BTRFS_INODE_ITEM_KEY
;
6303 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6304 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
6305 key
.objectid
, root
->root_key
.objectid
,
6306 parent_ino
, index
, name
, name_len
);
6307 } else if (add_backref
) {
6308 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
6312 /* Nothing to clean up yet */
6316 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
6318 btrfs_inode_type(inode
), index
);
6319 if (ret
== -EEXIST
|| ret
== -EOVERFLOW
)
6322 btrfs_abort_transaction(trans
, root
, ret
);
6326 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
6328 inode_inc_iversion(parent_inode
);
6329 parent_inode
->i_mtime
= parent_inode
->i_ctime
=
6330 current_fs_time(parent_inode
->i_sb
);
6331 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
6333 btrfs_abort_transaction(trans
, root
, ret
);
6337 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6340 err
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
6341 key
.objectid
, root
->root_key
.objectid
,
6342 parent_ino
, &local_index
, name
, name_len
);
6344 } else if (add_backref
) {
6348 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
6349 ino
, parent_ino
, &local_index
);
6354 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
6355 struct inode
*dir
, struct dentry
*dentry
,
6356 struct inode
*inode
, int backref
, u64 index
)
6358 int err
= btrfs_add_link(trans
, dir
, inode
,
6359 dentry
->d_name
.name
, dentry
->d_name
.len
,
6366 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
6367 umode_t mode
, dev_t rdev
)
6369 struct btrfs_trans_handle
*trans
;
6370 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6371 struct inode
*inode
= NULL
;
6378 * 2 for inode item and ref
6380 * 1 for xattr if selinux is on
6382 trans
= btrfs_start_transaction(root
, 5);
6384 return PTR_ERR(trans
);
6386 err
= btrfs_find_free_ino(root
, &objectid
);
6390 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6391 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6393 if (IS_ERR(inode
)) {
6394 err
= PTR_ERR(inode
);
6399 * If the active LSM wants to access the inode during
6400 * d_instantiate it needs these. Smack checks to see
6401 * if the filesystem supports xattrs by looking at the
6404 inode
->i_op
= &btrfs_special_inode_operations
;
6405 init_special_inode(inode
, inode
->i_mode
, rdev
);
6407 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6409 goto out_unlock_inode
;
6411 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
6413 goto out_unlock_inode
;
6415 btrfs_update_inode(trans
, root
, inode
);
6416 unlock_new_inode(inode
);
6417 d_instantiate(dentry
, inode
);
6421 btrfs_end_transaction(trans
, root
);
6422 btrfs_balance_delayed_items(root
);
6423 btrfs_btree_balance_dirty(root
);
6425 inode_dec_link_count(inode
);
6432 unlock_new_inode(inode
);
6437 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
6438 umode_t mode
, bool excl
)
6440 struct btrfs_trans_handle
*trans
;
6441 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6442 struct inode
*inode
= NULL
;
6443 int drop_inode_on_err
= 0;
6449 * 2 for inode item and ref
6451 * 1 for xattr if selinux is on
6453 trans
= btrfs_start_transaction(root
, 5);
6455 return PTR_ERR(trans
);
6457 err
= btrfs_find_free_ino(root
, &objectid
);
6461 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6462 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6464 if (IS_ERR(inode
)) {
6465 err
= PTR_ERR(inode
);
6468 drop_inode_on_err
= 1;
6470 * If the active LSM wants to access the inode during
6471 * d_instantiate it needs these. Smack checks to see
6472 * if the filesystem supports xattrs by looking at the
6475 inode
->i_fop
= &btrfs_file_operations
;
6476 inode
->i_op
= &btrfs_file_inode_operations
;
6477 inode
->i_mapping
->a_ops
= &btrfs_aops
;
6479 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6481 goto out_unlock_inode
;
6483 err
= btrfs_update_inode(trans
, root
, inode
);
6485 goto out_unlock_inode
;
6487 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
6489 goto out_unlock_inode
;
6491 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
6492 unlock_new_inode(inode
);
6493 d_instantiate(dentry
, inode
);
6496 btrfs_end_transaction(trans
, root
);
6497 if (err
&& drop_inode_on_err
) {
6498 inode_dec_link_count(inode
);
6501 btrfs_balance_delayed_items(root
);
6502 btrfs_btree_balance_dirty(root
);
6506 unlock_new_inode(inode
);
6511 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
6512 struct dentry
*dentry
)
6514 struct btrfs_trans_handle
*trans
= NULL
;
6515 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6516 struct inode
*inode
= d_inode(old_dentry
);
6521 /* do not allow sys_link's with other subvols of the same device */
6522 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
6525 if (inode
->i_nlink
>= BTRFS_LINK_MAX
)
6528 err
= btrfs_set_inode_index(dir
, &index
);
6533 * 2 items for inode and inode ref
6534 * 2 items for dir items
6535 * 1 item for parent inode
6537 trans
= btrfs_start_transaction(root
, 5);
6538 if (IS_ERR(trans
)) {
6539 err
= PTR_ERR(trans
);
6544 /* There are several dir indexes for this inode, clear the cache. */
6545 BTRFS_I(inode
)->dir_index
= 0ULL;
6547 inode_inc_iversion(inode
);
6548 inode
->i_ctime
= current_fs_time(inode
->i_sb
);
6550 set_bit(BTRFS_INODE_COPY_EVERYTHING
, &BTRFS_I(inode
)->runtime_flags
);
6552 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
6557 struct dentry
*parent
= dentry
->d_parent
;
6558 err
= btrfs_update_inode(trans
, root
, inode
);
6561 if (inode
->i_nlink
== 1) {
6563 * If new hard link count is 1, it's a file created
6564 * with open(2) O_TMPFILE flag.
6566 err
= btrfs_orphan_del(trans
, inode
);
6570 d_instantiate(dentry
, inode
);
6571 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
6574 btrfs_balance_delayed_items(root
);
6577 btrfs_end_transaction(trans
, root
);
6579 inode_dec_link_count(inode
);
6582 btrfs_btree_balance_dirty(root
);
6586 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
6588 struct inode
*inode
= NULL
;
6589 struct btrfs_trans_handle
*trans
;
6590 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6592 int drop_on_err
= 0;
6597 * 2 items for inode and ref
6598 * 2 items for dir items
6599 * 1 for xattr if selinux is on
6601 trans
= btrfs_start_transaction(root
, 5);
6603 return PTR_ERR(trans
);
6605 err
= btrfs_find_free_ino(root
, &objectid
);
6609 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6610 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6611 S_IFDIR
| mode
, &index
);
6612 if (IS_ERR(inode
)) {
6613 err
= PTR_ERR(inode
);
6618 /* these must be set before we unlock the inode */
6619 inode
->i_op
= &btrfs_dir_inode_operations
;
6620 inode
->i_fop
= &btrfs_dir_file_operations
;
6622 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6624 goto out_fail_inode
;
6626 btrfs_i_size_write(inode
, 0);
6627 err
= btrfs_update_inode(trans
, root
, inode
);
6629 goto out_fail_inode
;
6631 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
6632 dentry
->d_name
.len
, 0, index
);
6634 goto out_fail_inode
;
6636 d_instantiate(dentry
, inode
);
6638 * mkdir is special. We're unlocking after we call d_instantiate
6639 * to avoid a race with nfsd calling d_instantiate.
6641 unlock_new_inode(inode
);
6645 btrfs_end_transaction(trans
, root
);
6647 inode_dec_link_count(inode
);
6650 btrfs_balance_delayed_items(root
);
6651 btrfs_btree_balance_dirty(root
);
6655 unlock_new_inode(inode
);
6659 /* Find next extent map of a given extent map, caller needs to ensure locks */
6660 static struct extent_map
*next_extent_map(struct extent_map
*em
)
6662 struct rb_node
*next
;
6664 next
= rb_next(&em
->rb_node
);
6667 return container_of(next
, struct extent_map
, rb_node
);
6670 static struct extent_map
*prev_extent_map(struct extent_map
*em
)
6672 struct rb_node
*prev
;
6674 prev
= rb_prev(&em
->rb_node
);
6677 return container_of(prev
, struct extent_map
, rb_node
);
6680 /* helper for btfs_get_extent. Given an existing extent in the tree,
6681 * the existing extent is the nearest extent to map_start,
6682 * and an extent that you want to insert, deal with overlap and insert
6683 * the best fitted new extent into the tree.
6685 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
6686 struct extent_map
*existing
,
6687 struct extent_map
*em
,
6690 struct extent_map
*prev
;
6691 struct extent_map
*next
;
6696 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
6698 if (existing
->start
> map_start
) {
6700 prev
= prev_extent_map(next
);
6703 next
= next_extent_map(prev
);
6706 start
= prev
? extent_map_end(prev
) : em
->start
;
6707 start
= max_t(u64
, start
, em
->start
);
6708 end
= next
? next
->start
: extent_map_end(em
);
6709 end
= min_t(u64
, end
, extent_map_end(em
));
6710 start_diff
= start
- em
->start
;
6712 em
->len
= end
- start
;
6713 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
6714 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
6715 em
->block_start
+= start_diff
;
6716 em
->block_len
-= start_diff
;
6718 return add_extent_mapping(em_tree
, em
, 0);
6721 static noinline
int uncompress_inline(struct btrfs_path
*path
,
6723 size_t pg_offset
, u64 extent_offset
,
6724 struct btrfs_file_extent_item
*item
)
6727 struct extent_buffer
*leaf
= path
->nodes
[0];
6730 unsigned long inline_size
;
6734 WARN_ON(pg_offset
!= 0);
6735 compress_type
= btrfs_file_extent_compression(leaf
, item
);
6736 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
6737 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
6738 btrfs_item_nr(path
->slots
[0]));
6739 tmp
= kmalloc(inline_size
, GFP_NOFS
);
6742 ptr
= btrfs_file_extent_inline_start(item
);
6744 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
6746 max_size
= min_t(unsigned long, PAGE_SIZE
, max_size
);
6747 ret
= btrfs_decompress(compress_type
, tmp
, page
,
6748 extent_offset
, inline_size
, max_size
);
6754 * a bit scary, this does extent mapping from logical file offset to the disk.
6755 * the ugly parts come from merging extents from the disk with the in-ram
6756 * representation. This gets more complex because of the data=ordered code,
6757 * where the in-ram extents might be locked pending data=ordered completion.
6759 * This also copies inline extents directly into the page.
6762 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
6763 size_t pg_offset
, u64 start
, u64 len
,
6768 u64 extent_start
= 0;
6770 u64 objectid
= btrfs_ino(inode
);
6772 struct btrfs_path
*path
= NULL
;
6773 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6774 struct btrfs_file_extent_item
*item
;
6775 struct extent_buffer
*leaf
;
6776 struct btrfs_key found_key
;
6777 struct extent_map
*em
= NULL
;
6778 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
6779 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6780 struct btrfs_trans_handle
*trans
= NULL
;
6781 const bool new_inline
= !page
|| create
;
6784 read_lock(&em_tree
->lock
);
6785 em
= lookup_extent_mapping(em_tree
, start
, len
);
6787 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6788 read_unlock(&em_tree
->lock
);
6791 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
6792 free_extent_map(em
);
6793 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
6794 free_extent_map(em
);
6798 em
= alloc_extent_map();
6803 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6804 em
->start
= EXTENT_MAP_HOLE
;
6805 em
->orig_start
= EXTENT_MAP_HOLE
;
6807 em
->block_len
= (u64
)-1;
6810 path
= btrfs_alloc_path();
6816 * Chances are we'll be called again, so go ahead and do
6819 path
->reada
= READA_FORWARD
;
6822 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
6823 objectid
, start
, trans
!= NULL
);
6830 if (path
->slots
[0] == 0)
6835 leaf
= path
->nodes
[0];
6836 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6837 struct btrfs_file_extent_item
);
6838 /* are we inside the extent that was found? */
6839 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6840 found_type
= found_key
.type
;
6841 if (found_key
.objectid
!= objectid
||
6842 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
6844 * If we backup past the first extent we want to move forward
6845 * and see if there is an extent in front of us, otherwise we'll
6846 * say there is a hole for our whole search range which can
6853 found_type
= btrfs_file_extent_type(leaf
, item
);
6854 extent_start
= found_key
.offset
;
6855 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6856 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6857 extent_end
= extent_start
+
6858 btrfs_file_extent_num_bytes(leaf
, item
);
6859 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6861 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
6862 extent_end
= ALIGN(extent_start
+ size
, root
->sectorsize
);
6865 if (start
>= extent_end
) {
6867 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
6868 ret
= btrfs_next_leaf(root
, path
);
6875 leaf
= path
->nodes
[0];
6877 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6878 if (found_key
.objectid
!= objectid
||
6879 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6881 if (start
+ len
<= found_key
.offset
)
6883 if (start
> found_key
.offset
)
6886 em
->orig_start
= start
;
6887 em
->len
= found_key
.offset
- start
;
6891 btrfs_extent_item_to_extent_map(inode
, path
, item
, new_inline
, em
);
6893 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6894 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6896 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6900 size_t extent_offset
;
6906 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
6907 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
6908 copy_size
= min_t(u64
, PAGE_SIZE
- pg_offset
,
6909 size
- extent_offset
);
6910 em
->start
= extent_start
+ extent_offset
;
6911 em
->len
= ALIGN(copy_size
, root
->sectorsize
);
6912 em
->orig_block_len
= em
->len
;
6913 em
->orig_start
= em
->start
;
6914 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
6915 if (create
== 0 && !PageUptodate(page
)) {
6916 if (btrfs_file_extent_compression(leaf
, item
) !=
6917 BTRFS_COMPRESS_NONE
) {
6918 ret
= uncompress_inline(path
, page
, pg_offset
,
6919 extent_offset
, item
);
6926 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6928 if (pg_offset
+ copy_size
< PAGE_SIZE
) {
6929 memset(map
+ pg_offset
+ copy_size
, 0,
6930 PAGE_SIZE
- pg_offset
-
6935 flush_dcache_page(page
);
6936 } else if (create
&& PageUptodate(page
)) {
6940 free_extent_map(em
);
6943 btrfs_release_path(path
);
6944 trans
= btrfs_join_transaction(root
);
6947 return ERR_CAST(trans
);
6951 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6954 btrfs_mark_buffer_dirty(leaf
);
6956 set_extent_uptodate(io_tree
, em
->start
,
6957 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
6962 em
->orig_start
= start
;
6965 em
->block_start
= EXTENT_MAP_HOLE
;
6966 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
6968 btrfs_release_path(path
);
6969 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
6970 btrfs_err(root
->fs_info
, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6971 em
->start
, em
->len
, start
, len
);
6977 write_lock(&em_tree
->lock
);
6978 ret
= add_extent_mapping(em_tree
, em
, 0);
6979 /* it is possible that someone inserted the extent into the tree
6980 * while we had the lock dropped. It is also possible that
6981 * an overlapping map exists in the tree
6983 if (ret
== -EEXIST
) {
6984 struct extent_map
*existing
;
6988 existing
= search_extent_mapping(em_tree
, start
, len
);
6990 * existing will always be non-NULL, since there must be
6991 * extent causing the -EEXIST.
6993 if (existing
->start
== em
->start
&&
6994 extent_map_end(existing
) == extent_map_end(em
) &&
6995 em
->block_start
== existing
->block_start
) {
6997 * these two extents are the same, it happens
6998 * with inlines especially
7000 free_extent_map(em
);
7004 } else if (start
>= extent_map_end(existing
) ||
7005 start
<= existing
->start
) {
7007 * The existing extent map is the one nearest to
7008 * the [start, start + len) range which overlaps
7010 err
= merge_extent_mapping(em_tree
, existing
,
7012 free_extent_map(existing
);
7014 free_extent_map(em
);
7018 free_extent_map(em
);
7023 write_unlock(&em_tree
->lock
);
7026 trace_btrfs_get_extent(root
, em
);
7028 btrfs_free_path(path
);
7030 ret
= btrfs_end_transaction(trans
, root
);
7035 free_extent_map(em
);
7036 return ERR_PTR(err
);
7038 BUG_ON(!em
); /* Error is always set */
7042 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
7043 size_t pg_offset
, u64 start
, u64 len
,
7046 struct extent_map
*em
;
7047 struct extent_map
*hole_em
= NULL
;
7048 u64 range_start
= start
;
7054 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
7061 * - a pre-alloc extent,
7062 * there might actually be delalloc bytes behind it.
7064 if (em
->block_start
!= EXTENT_MAP_HOLE
&&
7065 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7071 /* check to see if we've wrapped (len == -1 or similar) */
7080 /* ok, we didn't find anything, lets look for delalloc */
7081 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
7082 end
, len
, EXTENT_DELALLOC
, 1);
7083 found_end
= range_start
+ found
;
7084 if (found_end
< range_start
)
7085 found_end
= (u64
)-1;
7088 * we didn't find anything useful, return
7089 * the original results from get_extent()
7091 if (range_start
> end
|| found_end
<= start
) {
7097 /* adjust the range_start to make sure it doesn't
7098 * go backwards from the start they passed in
7100 range_start
= max(start
, range_start
);
7101 found
= found_end
- range_start
;
7104 u64 hole_start
= start
;
7107 em
= alloc_extent_map();
7113 * when btrfs_get_extent can't find anything it
7114 * returns one huge hole
7116 * make sure what it found really fits our range, and
7117 * adjust to make sure it is based on the start from
7121 u64 calc_end
= extent_map_end(hole_em
);
7123 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
7124 free_extent_map(hole_em
);
7127 hole_start
= max(hole_em
->start
, start
);
7128 hole_len
= calc_end
- hole_start
;
7132 if (hole_em
&& range_start
> hole_start
) {
7133 /* our hole starts before our delalloc, so we
7134 * have to return just the parts of the hole
7135 * that go until the delalloc starts
7137 em
->len
= min(hole_len
,
7138 range_start
- hole_start
);
7139 em
->start
= hole_start
;
7140 em
->orig_start
= hole_start
;
7142 * don't adjust block start at all,
7143 * it is fixed at EXTENT_MAP_HOLE
7145 em
->block_start
= hole_em
->block_start
;
7146 em
->block_len
= hole_len
;
7147 if (test_bit(EXTENT_FLAG_PREALLOC
, &hole_em
->flags
))
7148 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
7150 em
->start
= range_start
;
7152 em
->orig_start
= range_start
;
7153 em
->block_start
= EXTENT_MAP_DELALLOC
;
7154 em
->block_len
= found
;
7156 } else if (hole_em
) {
7161 free_extent_map(hole_em
);
7163 free_extent_map(em
);
7164 return ERR_PTR(err
);
7169 static struct extent_map
*btrfs_create_dio_extent(struct inode
*inode
,
7172 const u64 orig_start
,
7173 const u64 block_start
,
7174 const u64 block_len
,
7175 const u64 orig_block_len
,
7176 const u64 ram_bytes
,
7179 struct extent_map
*em
= NULL
;
7182 down_read(&BTRFS_I(inode
)->dio_sem
);
7183 if (type
!= BTRFS_ORDERED_NOCOW
) {
7184 em
= create_pinned_em(inode
, start
, len
, orig_start
,
7185 block_start
, block_len
, orig_block_len
,
7190 ret
= btrfs_add_ordered_extent_dio(inode
, start
, block_start
,
7191 len
, block_len
, type
);
7194 free_extent_map(em
);
7195 btrfs_drop_extent_cache(inode
, start
,
7196 start
+ len
- 1, 0);
7201 up_read(&BTRFS_I(inode
)->dio_sem
);
7206 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
7209 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7210 struct extent_map
*em
;
7211 struct btrfs_key ins
;
7215 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
7216 ret
= btrfs_reserve_extent(root
, len
, root
->sectorsize
, 0,
7217 alloc_hint
, &ins
, 1, 1);
7219 return ERR_PTR(ret
);
7221 em
= btrfs_create_dio_extent(inode
, start
, ins
.offset
, start
,
7222 ins
.objectid
, ins
.offset
, ins
.offset
,
7224 btrfs_dec_block_group_reservations(root
->fs_info
, ins
.objectid
);
7226 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
7232 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7233 * block must be cow'd
7235 noinline
int can_nocow_extent(struct inode
*inode
, u64 offset
, u64
*len
,
7236 u64
*orig_start
, u64
*orig_block_len
,
7239 struct btrfs_trans_handle
*trans
;
7240 struct btrfs_path
*path
;
7242 struct extent_buffer
*leaf
;
7243 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7244 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7245 struct btrfs_file_extent_item
*fi
;
7246 struct btrfs_key key
;
7253 bool nocow
= (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
);
7255 path
= btrfs_alloc_path();
7259 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, btrfs_ino(inode
),
7264 slot
= path
->slots
[0];
7267 /* can't find the item, must cow */
7274 leaf
= path
->nodes
[0];
7275 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
7276 if (key
.objectid
!= btrfs_ino(inode
) ||
7277 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
7278 /* not our file or wrong item type, must cow */
7282 if (key
.offset
> offset
) {
7283 /* Wrong offset, must cow */
7287 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
7288 found_type
= btrfs_file_extent_type(leaf
, fi
);
7289 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
7290 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
7291 /* not a regular extent, must cow */
7295 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_REG
)
7298 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
7299 if (extent_end
<= offset
)
7302 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
7303 if (disk_bytenr
== 0)
7306 if (btrfs_file_extent_compression(leaf
, fi
) ||
7307 btrfs_file_extent_encryption(leaf
, fi
) ||
7308 btrfs_file_extent_other_encoding(leaf
, fi
))
7311 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
7314 *orig_start
= key
.offset
- backref_offset
;
7315 *orig_block_len
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7316 *ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
7319 if (btrfs_extent_readonly(root
, disk_bytenr
))
7322 num_bytes
= min(offset
+ *len
, extent_end
) - offset
;
7323 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7326 range_end
= round_up(offset
+ num_bytes
, root
->sectorsize
) - 1;
7327 ret
= test_range_bit(io_tree
, offset
, range_end
,
7328 EXTENT_DELALLOC
, 0, NULL
);
7335 btrfs_release_path(path
);
7338 * look for other files referencing this extent, if we
7339 * find any we must cow
7341 trans
= btrfs_join_transaction(root
);
7342 if (IS_ERR(trans
)) {
7347 ret
= btrfs_cross_ref_exist(trans
, root
, btrfs_ino(inode
),
7348 key
.offset
- backref_offset
, disk_bytenr
);
7349 btrfs_end_transaction(trans
, root
);
7356 * adjust disk_bytenr and num_bytes to cover just the bytes
7357 * in this extent we are about to write. If there
7358 * are any csums in that range we have to cow in order
7359 * to keep the csums correct
7361 disk_bytenr
+= backref_offset
;
7362 disk_bytenr
+= offset
- key
.offset
;
7363 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
7366 * all of the above have passed, it is safe to overwrite this extent
7372 btrfs_free_path(path
);
7376 bool btrfs_page_exists_in_range(struct inode
*inode
, loff_t start
, loff_t end
)
7378 struct radix_tree_root
*root
= &inode
->i_mapping
->page_tree
;
7380 void **pagep
= NULL
;
7381 struct page
*page
= NULL
;
7385 start_idx
= start
>> PAGE_SHIFT
;
7388 * end is the last byte in the last page. end == start is legal
7390 end_idx
= end
>> PAGE_SHIFT
;
7394 /* Most of the code in this while loop is lifted from
7395 * find_get_page. It's been modified to begin searching from a
7396 * page and return just the first page found in that range. If the
7397 * found idx is less than or equal to the end idx then we know that
7398 * a page exists. If no pages are found or if those pages are
7399 * outside of the range then we're fine (yay!) */
7400 while (page
== NULL
&&
7401 radix_tree_gang_lookup_slot(root
, &pagep
, NULL
, start_idx
, 1)) {
7402 page
= radix_tree_deref_slot(pagep
);
7403 if (unlikely(!page
))
7406 if (radix_tree_exception(page
)) {
7407 if (radix_tree_deref_retry(page
)) {
7412 * Otherwise, shmem/tmpfs must be storing a swap entry
7413 * here as an exceptional entry: so return it without
7414 * attempting to raise page count.
7417 break; /* TODO: Is this relevant for this use case? */
7420 if (!page_cache_get_speculative(page
)) {
7426 * Has the page moved?
7427 * This is part of the lockless pagecache protocol. See
7428 * include/linux/pagemap.h for details.
7430 if (unlikely(page
!= *pagep
)) {
7437 if (page
->index
<= end_idx
)
7446 static int lock_extent_direct(struct inode
*inode
, u64 lockstart
, u64 lockend
,
7447 struct extent_state
**cached_state
, int writing
)
7449 struct btrfs_ordered_extent
*ordered
;
7453 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7456 * We're concerned with the entire range that we're going to be
7457 * doing DIO to, so we need to make sure there's no ordered
7458 * extents in this range.
7460 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
7461 lockend
- lockstart
+ 1);
7464 * We need to make sure there are no buffered pages in this
7465 * range either, we could have raced between the invalidate in
7466 * generic_file_direct_write and locking the extent. The
7467 * invalidate needs to happen so that reads after a write do not
7472 !btrfs_page_exists_in_range(inode
, lockstart
, lockend
)))
7475 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7476 cached_state
, GFP_NOFS
);
7480 * If we are doing a DIO read and the ordered extent we
7481 * found is for a buffered write, we can not wait for it
7482 * to complete and retry, because if we do so we can
7483 * deadlock with concurrent buffered writes on page
7484 * locks. This happens only if our DIO read covers more
7485 * than one extent map, if at this point has already
7486 * created an ordered extent for a previous extent map
7487 * and locked its range in the inode's io tree, and a
7488 * concurrent write against that previous extent map's
7489 * range and this range started (we unlock the ranges
7490 * in the io tree only when the bios complete and
7491 * buffered writes always lock pages before attempting
7492 * to lock range in the io tree).
7495 test_bit(BTRFS_ORDERED_DIRECT
, &ordered
->flags
))
7496 btrfs_start_ordered_extent(inode
, ordered
, 1);
7499 btrfs_put_ordered_extent(ordered
);
7502 * We could trigger writeback for this range (and wait
7503 * for it to complete) and then invalidate the pages for
7504 * this range (through invalidate_inode_pages2_range()),
7505 * but that can lead us to a deadlock with a concurrent
7506 * call to readpages() (a buffered read or a defrag call
7507 * triggered a readahead) on a page lock due to an
7508 * ordered dio extent we created before but did not have
7509 * yet a corresponding bio submitted (whence it can not
7510 * complete), which makes readpages() wait for that
7511 * ordered extent to complete while holding a lock on
7526 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
7527 u64 len
, u64 orig_start
,
7528 u64 block_start
, u64 block_len
,
7529 u64 orig_block_len
, u64 ram_bytes
,
7532 struct extent_map_tree
*em_tree
;
7533 struct extent_map
*em
;
7534 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7537 em_tree
= &BTRFS_I(inode
)->extent_tree
;
7538 em
= alloc_extent_map();
7540 return ERR_PTR(-ENOMEM
);
7543 em
->orig_start
= orig_start
;
7544 em
->mod_start
= start
;
7547 em
->block_len
= block_len
;
7548 em
->block_start
= block_start
;
7549 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
7550 em
->orig_block_len
= orig_block_len
;
7551 em
->ram_bytes
= ram_bytes
;
7552 em
->generation
= -1;
7553 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
7554 if (type
== BTRFS_ORDERED_PREALLOC
)
7555 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
7558 btrfs_drop_extent_cache(inode
, em
->start
,
7559 em
->start
+ em
->len
- 1, 0);
7560 write_lock(&em_tree
->lock
);
7561 ret
= add_extent_mapping(em_tree
, em
, 1);
7562 write_unlock(&em_tree
->lock
);
7563 } while (ret
== -EEXIST
);
7566 free_extent_map(em
);
7567 return ERR_PTR(ret
);
7573 static void adjust_dio_outstanding_extents(struct inode
*inode
,
7574 struct btrfs_dio_data
*dio_data
,
7577 unsigned num_extents
;
7579 num_extents
= (unsigned) div64_u64(len
+ BTRFS_MAX_EXTENT_SIZE
- 1,
7580 BTRFS_MAX_EXTENT_SIZE
);
7582 * If we have an outstanding_extents count still set then we're
7583 * within our reservation, otherwise we need to adjust our inode
7584 * counter appropriately.
7586 if (dio_data
->outstanding_extents
) {
7587 dio_data
->outstanding_extents
-= num_extents
;
7589 spin_lock(&BTRFS_I(inode
)->lock
);
7590 BTRFS_I(inode
)->outstanding_extents
+= num_extents
;
7591 spin_unlock(&BTRFS_I(inode
)->lock
);
7595 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
7596 struct buffer_head
*bh_result
, int create
)
7598 struct extent_map
*em
;
7599 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7600 struct extent_state
*cached_state
= NULL
;
7601 struct btrfs_dio_data
*dio_data
= NULL
;
7602 u64 start
= iblock
<< inode
->i_blkbits
;
7603 u64 lockstart
, lockend
;
7604 u64 len
= bh_result
->b_size
;
7605 int unlock_bits
= EXTENT_LOCKED
;
7609 unlock_bits
|= EXTENT_DIRTY
;
7611 len
= min_t(u64
, len
, root
->sectorsize
);
7614 lockend
= start
+ len
- 1;
7616 if (current
->journal_info
) {
7618 * Need to pull our outstanding extents and set journal_info to NULL so
7619 * that anything that needs to check if there's a transaction doesn't get
7622 dio_data
= current
->journal_info
;
7623 current
->journal_info
= NULL
;
7627 * If this errors out it's because we couldn't invalidate pagecache for
7628 * this range and we need to fallback to buffered.
7630 if (lock_extent_direct(inode
, lockstart
, lockend
, &cached_state
,
7636 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
7643 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7644 * io. INLINE is special, and we could probably kludge it in here, but
7645 * it's still buffered so for safety lets just fall back to the generic
7648 * For COMPRESSED we _have_ to read the entire extent in so we can
7649 * decompress it, so there will be buffering required no matter what we
7650 * do, so go ahead and fallback to buffered.
7652 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7653 * to buffered IO. Don't blame me, this is the price we pay for using
7656 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
7657 em
->block_start
== EXTENT_MAP_INLINE
) {
7658 free_extent_map(em
);
7663 /* Just a good old fashioned hole, return */
7664 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
7665 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
7666 free_extent_map(em
);
7671 * We don't allocate a new extent in the following cases
7673 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7675 * 2) The extent is marked as PREALLOC. We're good to go here and can
7676 * just use the extent.
7680 len
= min(len
, em
->len
- (start
- em
->start
));
7681 lockstart
= start
+ len
;
7685 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
7686 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
7687 em
->block_start
!= EXTENT_MAP_HOLE
)) {
7689 u64 block_start
, orig_start
, orig_block_len
, ram_bytes
;
7691 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7692 type
= BTRFS_ORDERED_PREALLOC
;
7694 type
= BTRFS_ORDERED_NOCOW
;
7695 len
= min(len
, em
->len
- (start
- em
->start
));
7696 block_start
= em
->block_start
+ (start
- em
->start
);
7698 if (can_nocow_extent(inode
, start
, &len
, &orig_start
,
7699 &orig_block_len
, &ram_bytes
) == 1 &&
7700 btrfs_inc_nocow_writers(root
->fs_info
, block_start
)) {
7701 struct extent_map
*em2
;
7703 em2
= btrfs_create_dio_extent(inode
, start
, len
,
7704 orig_start
, block_start
,
7705 len
, orig_block_len
,
7707 btrfs_dec_nocow_writers(root
->fs_info
, block_start
);
7708 if (type
== BTRFS_ORDERED_PREALLOC
) {
7709 free_extent_map(em
);
7712 if (em2
&& IS_ERR(em2
)) {
7721 * this will cow the extent, reset the len in case we changed
7724 len
= bh_result
->b_size
;
7725 free_extent_map(em
);
7726 em
= btrfs_new_extent_direct(inode
, start
, len
);
7731 len
= min(len
, em
->len
- (start
- em
->start
));
7733 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
7735 bh_result
->b_size
= len
;
7736 bh_result
->b_bdev
= em
->bdev
;
7737 set_buffer_mapped(bh_result
);
7739 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7740 set_buffer_new(bh_result
);
7743 * Need to update the i_size under the extent lock so buffered
7744 * readers will get the updated i_size when we unlock.
7746 if (start
+ len
> i_size_read(inode
))
7747 i_size_write(inode
, start
+ len
);
7749 adjust_dio_outstanding_extents(inode
, dio_data
, len
);
7750 btrfs_free_reserved_data_space(inode
, start
, len
);
7751 WARN_ON(dio_data
->reserve
< len
);
7752 dio_data
->reserve
-= len
;
7753 dio_data
->unsubmitted_oe_range_end
= start
+ len
;
7754 current
->journal_info
= dio_data
;
7758 * In the case of write we need to clear and unlock the entire range,
7759 * in the case of read we need to unlock only the end area that we
7760 * aren't using if there is any left over space.
7762 if (lockstart
< lockend
) {
7763 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
7764 lockend
, unlock_bits
, 1, 0,
7765 &cached_state
, GFP_NOFS
);
7767 free_extent_state(cached_state
);
7770 free_extent_map(em
);
7775 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7776 unlock_bits
, 1, 0, &cached_state
, GFP_NOFS
);
7779 current
->journal_info
= dio_data
;
7781 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7782 * write less data then expected, so that we don't underflow our inode's
7783 * outstanding extents counter.
7785 if (create
&& dio_data
)
7786 adjust_dio_outstanding_extents(inode
, dio_data
, len
);
7791 static inline int submit_dio_repair_bio(struct inode
*inode
, struct bio
*bio
,
7794 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7797 BUG_ON(bio_op(bio
) == REQ_OP_WRITE
);
7801 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
,
7802 BTRFS_WQ_ENDIO_DIO_REPAIR
);
7806 ret
= btrfs_map_bio(root
, bio
, mirror_num
, 0);
7812 static int btrfs_check_dio_repairable(struct inode
*inode
,
7813 struct bio
*failed_bio
,
7814 struct io_failure_record
*failrec
,
7819 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
7820 failrec
->logical
, failrec
->len
);
7821 if (num_copies
== 1) {
7823 * we only have a single copy of the data, so don't bother with
7824 * all the retry and error correction code that follows. no
7825 * matter what the error is, it is very likely to persist.
7827 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7828 num_copies
, failrec
->this_mirror
, failed_mirror
);
7832 failrec
->failed_mirror
= failed_mirror
;
7833 failrec
->this_mirror
++;
7834 if (failrec
->this_mirror
== failed_mirror
)
7835 failrec
->this_mirror
++;
7837 if (failrec
->this_mirror
> num_copies
) {
7838 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7839 num_copies
, failrec
->this_mirror
, failed_mirror
);
7846 static int dio_read_error(struct inode
*inode
, struct bio
*failed_bio
,
7847 struct page
*page
, unsigned int pgoff
,
7848 u64 start
, u64 end
, int failed_mirror
,
7849 bio_end_io_t
*repair_endio
, void *repair_arg
)
7851 struct io_failure_record
*failrec
;
7857 BUG_ON(bio_op(failed_bio
) == REQ_OP_WRITE
);
7859 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
7863 ret
= btrfs_check_dio_repairable(inode
, failed_bio
, failrec
,
7866 free_io_failure(inode
, failrec
);
7870 if ((failed_bio
->bi_vcnt
> 1)
7871 || (failed_bio
->bi_io_vec
->bv_len
7872 > BTRFS_I(inode
)->root
->sectorsize
))
7873 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
7875 read_mode
= READ_SYNC
;
7877 isector
= start
- btrfs_io_bio(failed_bio
)->logical
;
7878 isector
>>= inode
->i_sb
->s_blocksize_bits
;
7879 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
7880 pgoff
, isector
, repair_endio
, repair_arg
);
7882 free_io_failure(inode
, failrec
);
7885 bio_set_op_attrs(bio
, REQ_OP_READ
, read_mode
);
7887 btrfs_debug(BTRFS_I(inode
)->root
->fs_info
,
7888 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7889 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
7891 ret
= submit_dio_repair_bio(inode
, bio
, failrec
->this_mirror
);
7893 free_io_failure(inode
, failrec
);
7900 struct btrfs_retry_complete
{
7901 struct completion done
;
7902 struct inode
*inode
;
7907 static void btrfs_retry_endio_nocsum(struct bio
*bio
)
7909 struct btrfs_retry_complete
*done
= bio
->bi_private
;
7910 struct inode
*inode
;
7911 struct bio_vec
*bvec
;
7917 ASSERT(bio
->bi_vcnt
== 1);
7918 inode
= bio
->bi_io_vec
->bv_page
->mapping
->host
;
7919 ASSERT(bio
->bi_io_vec
->bv_len
== BTRFS_I(inode
)->root
->sectorsize
);
7922 bio_for_each_segment_all(bvec
, bio
, i
)
7923 clean_io_failure(done
->inode
, done
->start
, bvec
->bv_page
, 0);
7925 complete(&done
->done
);
7929 static int __btrfs_correct_data_nocsum(struct inode
*inode
,
7930 struct btrfs_io_bio
*io_bio
)
7932 struct btrfs_fs_info
*fs_info
;
7933 struct bio_vec
*bvec
;
7934 struct btrfs_retry_complete done
;
7942 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
7943 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
7945 start
= io_bio
->logical
;
7948 bio_for_each_segment_all(bvec
, &io_bio
->bio
, i
) {
7949 nr_sectors
= BTRFS_BYTES_TO_BLKS(fs_info
, bvec
->bv_len
);
7950 pgoff
= bvec
->bv_offset
;
7952 next_block_or_try_again
:
7955 init_completion(&done
.done
);
7957 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
->bv_page
,
7958 pgoff
, start
, start
+ sectorsize
- 1,
7960 btrfs_retry_endio_nocsum
, &done
);
7964 wait_for_completion(&done
.done
);
7966 if (!done
.uptodate
) {
7967 /* We might have another mirror, so try again */
7968 goto next_block_or_try_again
;
7971 start
+= sectorsize
;
7974 pgoff
+= sectorsize
;
7975 goto next_block_or_try_again
;
7982 static void btrfs_retry_endio(struct bio
*bio
)
7984 struct btrfs_retry_complete
*done
= bio
->bi_private
;
7985 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
7986 struct inode
*inode
;
7987 struct bio_vec
*bvec
;
7998 start
= done
->start
;
8000 ASSERT(bio
->bi_vcnt
== 1);
8001 inode
= bio
->bi_io_vec
->bv_page
->mapping
->host
;
8002 ASSERT(bio
->bi_io_vec
->bv_len
== BTRFS_I(inode
)->root
->sectorsize
);
8004 bio_for_each_segment_all(bvec
, bio
, i
) {
8005 ret
= __readpage_endio_check(done
->inode
, io_bio
, i
,
8006 bvec
->bv_page
, bvec
->bv_offset
,
8007 done
->start
, bvec
->bv_len
);
8009 clean_io_failure(done
->inode
, done
->start
,
8010 bvec
->bv_page
, bvec
->bv_offset
);
8015 done
->uptodate
= uptodate
;
8017 complete(&done
->done
);
8021 static int __btrfs_subio_endio_read(struct inode
*inode
,
8022 struct btrfs_io_bio
*io_bio
, int err
)
8024 struct btrfs_fs_info
*fs_info
;
8025 struct bio_vec
*bvec
;
8026 struct btrfs_retry_complete done
;
8036 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
8037 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
8040 start
= io_bio
->logical
;
8043 bio_for_each_segment_all(bvec
, &io_bio
->bio
, i
) {
8044 nr_sectors
= BTRFS_BYTES_TO_BLKS(fs_info
, bvec
->bv_len
);
8046 pgoff
= bvec
->bv_offset
;
8048 csum_pos
= BTRFS_BYTES_TO_BLKS(fs_info
, offset
);
8049 ret
= __readpage_endio_check(inode
, io_bio
, csum_pos
,
8050 bvec
->bv_page
, pgoff
, start
,
8057 init_completion(&done
.done
);
8059 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
->bv_page
,
8060 pgoff
, start
, start
+ sectorsize
- 1,
8062 btrfs_retry_endio
, &done
);
8068 wait_for_completion(&done
.done
);
8070 if (!done
.uptodate
) {
8071 /* We might have another mirror, so try again */
8075 offset
+= sectorsize
;
8076 start
+= sectorsize
;
8081 pgoff
+= sectorsize
;
8089 static int btrfs_subio_endio_read(struct inode
*inode
,
8090 struct btrfs_io_bio
*io_bio
, int err
)
8092 bool skip_csum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
8096 return __btrfs_correct_data_nocsum(inode
, io_bio
);
8100 return __btrfs_subio_endio_read(inode
, io_bio
, err
);
8104 static void btrfs_endio_direct_read(struct bio
*bio
)
8106 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8107 struct inode
*inode
= dip
->inode
;
8108 struct bio
*dio_bio
;
8109 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8110 int err
= bio
->bi_error
;
8112 if (dip
->flags
& BTRFS_DIO_ORIG_BIO_SUBMITTED
)
8113 err
= btrfs_subio_endio_read(inode
, io_bio
, err
);
8115 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
8116 dip
->logical_offset
+ dip
->bytes
- 1);
8117 dio_bio
= dip
->dio_bio
;
8121 dio_bio
->bi_error
= bio
->bi_error
;
8122 dio_end_io(dio_bio
, bio
->bi_error
);
8125 io_bio
->end_io(io_bio
, err
);
8129 static void btrfs_endio_direct_write_update_ordered(struct inode
*inode
,
8134 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8135 struct btrfs_ordered_extent
*ordered
= NULL
;
8136 u64 ordered_offset
= offset
;
8137 u64 ordered_bytes
= bytes
;
8141 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
8148 btrfs_init_work(&ordered
->work
, btrfs_endio_write_helper
,
8149 finish_ordered_fn
, NULL
, NULL
);
8150 btrfs_queue_work(root
->fs_info
->endio_write_workers
,
8154 * our bio might span multiple ordered extents. If we haven't
8155 * completed the accounting for the whole dio, go back and try again
8157 if (ordered_offset
< offset
+ bytes
) {
8158 ordered_bytes
= offset
+ bytes
- ordered_offset
;
8164 static void btrfs_endio_direct_write(struct bio
*bio
)
8166 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8167 struct bio
*dio_bio
= dip
->dio_bio
;
8169 btrfs_endio_direct_write_update_ordered(dip
->inode
,
8170 dip
->logical_offset
,
8176 dio_bio
->bi_error
= bio
->bi_error
;
8177 dio_end_io(dio_bio
, bio
->bi_error
);
8181 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
,
8182 struct bio
*bio
, int mirror_num
,
8183 unsigned long bio_flags
, u64 offset
)
8186 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8187 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
8188 BUG_ON(ret
); /* -ENOMEM */
8192 static void btrfs_end_dio_bio(struct bio
*bio
)
8194 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8195 int err
= bio
->bi_error
;
8198 btrfs_warn(BTRFS_I(dip
->inode
)->root
->fs_info
,
8199 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8200 btrfs_ino(dip
->inode
), bio_op(bio
), bio
->bi_rw
,
8201 (unsigned long long)bio
->bi_iter
.bi_sector
,
8202 bio
->bi_iter
.bi_size
, err
);
8204 if (dip
->subio_endio
)
8205 err
= dip
->subio_endio(dip
->inode
, btrfs_io_bio(bio
), err
);
8211 * before atomic variable goto zero, we must make sure
8212 * dip->errors is perceived to be set.
8214 smp_mb__before_atomic();
8217 /* if there are more bios still pending for this dio, just exit */
8218 if (!atomic_dec_and_test(&dip
->pending_bios
))
8222 bio_io_error(dip
->orig_bio
);
8224 dip
->dio_bio
->bi_error
= 0;
8225 bio_endio(dip
->orig_bio
);
8231 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
8232 u64 first_sector
, gfp_t gfp_flags
)
8235 bio
= btrfs_bio_alloc(bdev
, first_sector
, BIO_MAX_PAGES
, gfp_flags
);
8237 bio_associate_current(bio
);
8241 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root
*root
,
8242 struct inode
*inode
,
8243 struct btrfs_dio_private
*dip
,
8247 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8248 struct btrfs_io_bio
*orig_io_bio
= btrfs_io_bio(dip
->orig_bio
);
8252 * We load all the csum data we need when we submit
8253 * the first bio to reduce the csum tree search and
8256 if (dip
->logical_offset
== file_offset
) {
8257 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, dip
->orig_bio
,
8263 if (bio
== dip
->orig_bio
)
8266 file_offset
-= dip
->logical_offset
;
8267 file_offset
>>= inode
->i_sb
->s_blocksize_bits
;
8268 io_bio
->csum
= (u8
*)(((u32
*)orig_io_bio
->csum
) + file_offset
);
8273 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
8274 u64 file_offset
, int skip_sum
,
8277 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8278 bool write
= bio_op(bio
) == REQ_OP_WRITE
;
8279 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8283 async_submit
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
8288 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
,
8289 BTRFS_WQ_ENDIO_DATA
);
8297 if (write
&& async_submit
) {
8298 ret
= btrfs_wq_submit_bio(root
->fs_info
,
8299 inode
, bio
, 0, 0, file_offset
,
8300 __btrfs_submit_bio_start_direct_io
,
8301 __btrfs_submit_bio_done
);
8305 * If we aren't doing async submit, calculate the csum of the
8308 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
8312 ret
= btrfs_lookup_and_bind_dio_csum(root
, inode
, dip
, bio
,
8318 ret
= btrfs_map_bio(root
, bio
, 0, async_submit
);
8324 static int btrfs_submit_direct_hook(struct btrfs_dio_private
*dip
,
8327 struct inode
*inode
= dip
->inode
;
8328 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8330 struct bio
*orig_bio
= dip
->orig_bio
;
8331 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
8332 u64 start_sector
= orig_bio
->bi_iter
.bi_sector
;
8333 u64 file_offset
= dip
->logical_offset
;
8336 u32 blocksize
= root
->sectorsize
;
8337 int async_submit
= 0;
8342 map_length
= orig_bio
->bi_iter
.bi_size
;
8343 ret
= btrfs_map_block(root
->fs_info
, bio_op(orig_bio
),
8344 start_sector
<< 9, &map_length
, NULL
, 0);
8348 if (map_length
>= orig_bio
->bi_iter
.bi_size
) {
8350 dip
->flags
|= BTRFS_DIO_ORIG_BIO_SUBMITTED
;
8354 /* async crcs make it difficult to collect full stripe writes. */
8355 if (btrfs_get_alloc_profile(root
, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK
)
8360 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
8364 bio_set_op_attrs(bio
, bio_op(orig_bio
), orig_bio
->bi_rw
);
8365 bio
->bi_private
= dip
;
8366 bio
->bi_end_io
= btrfs_end_dio_bio
;
8367 btrfs_io_bio(bio
)->logical
= file_offset
;
8368 atomic_inc(&dip
->pending_bios
);
8370 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
8371 nr_sectors
= BTRFS_BYTES_TO_BLKS(root
->fs_info
, bvec
->bv_len
);
8374 if (unlikely(map_length
< submit_len
+ blocksize
||
8375 bio_add_page(bio
, bvec
->bv_page
, blocksize
,
8376 bvec
->bv_offset
+ (i
* blocksize
)) < blocksize
)) {
8378 * inc the count before we submit the bio so
8379 * we know the end IO handler won't happen before
8380 * we inc the count. Otherwise, the dip might get freed
8381 * before we're done setting it up
8383 atomic_inc(&dip
->pending_bios
);
8384 ret
= __btrfs_submit_dio_bio(bio
, inode
,
8385 file_offset
, skip_sum
,
8389 atomic_dec(&dip
->pending_bios
);
8393 start_sector
+= submit_len
>> 9;
8394 file_offset
+= submit_len
;
8398 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
8399 start_sector
, GFP_NOFS
);
8402 bio_set_op_attrs(bio
, bio_op(orig_bio
), orig_bio
->bi_rw
);
8403 bio
->bi_private
= dip
;
8404 bio
->bi_end_io
= btrfs_end_dio_bio
;
8405 btrfs_io_bio(bio
)->logical
= file_offset
;
8407 map_length
= orig_bio
->bi_iter
.bi_size
;
8408 ret
= btrfs_map_block(root
->fs_info
, bio_op(orig_bio
),
8410 &map_length
, NULL
, 0);
8418 submit_len
+= blocksize
;
8428 ret
= __btrfs_submit_dio_bio(bio
, inode
, file_offset
, skip_sum
,
8437 * before atomic variable goto zero, we must
8438 * make sure dip->errors is perceived to be set.
8440 smp_mb__before_atomic();
8441 if (atomic_dec_and_test(&dip
->pending_bios
))
8442 bio_io_error(dip
->orig_bio
);
8444 /* bio_end_io() will handle error, so we needn't return it */
8448 static void btrfs_submit_direct(struct bio
*dio_bio
, struct inode
*inode
,
8451 struct btrfs_dio_private
*dip
= NULL
;
8452 struct bio
*io_bio
= NULL
;
8453 struct btrfs_io_bio
*btrfs_bio
;
8455 bool write
= (bio_op(dio_bio
) == REQ_OP_WRITE
);
8458 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
8460 io_bio
= btrfs_bio_clone(dio_bio
, GFP_NOFS
);
8466 dip
= kzalloc(sizeof(*dip
), GFP_NOFS
);
8472 dip
->private = dio_bio
->bi_private
;
8474 dip
->logical_offset
= file_offset
;
8475 dip
->bytes
= dio_bio
->bi_iter
.bi_size
;
8476 dip
->disk_bytenr
= (u64
)dio_bio
->bi_iter
.bi_sector
<< 9;
8477 io_bio
->bi_private
= dip
;
8478 dip
->orig_bio
= io_bio
;
8479 dip
->dio_bio
= dio_bio
;
8480 atomic_set(&dip
->pending_bios
, 0);
8481 btrfs_bio
= btrfs_io_bio(io_bio
);
8482 btrfs_bio
->logical
= file_offset
;
8485 io_bio
->bi_end_io
= btrfs_endio_direct_write
;
8487 io_bio
->bi_end_io
= btrfs_endio_direct_read
;
8488 dip
->subio_endio
= btrfs_subio_endio_read
;
8492 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8493 * even if we fail to submit a bio, because in such case we do the
8494 * corresponding error handling below and it must not be done a second
8495 * time by btrfs_direct_IO().
8498 struct btrfs_dio_data
*dio_data
= current
->journal_info
;
8500 dio_data
->unsubmitted_oe_range_end
= dip
->logical_offset
+
8502 dio_data
->unsubmitted_oe_range_start
=
8503 dio_data
->unsubmitted_oe_range_end
;
8506 ret
= btrfs_submit_direct_hook(dip
, skip_sum
);
8510 if (btrfs_bio
->end_io
)
8511 btrfs_bio
->end_io(btrfs_bio
, ret
);
8515 * If we arrived here it means either we failed to submit the dip
8516 * or we either failed to clone the dio_bio or failed to allocate the
8517 * dip. If we cloned the dio_bio and allocated the dip, we can just
8518 * call bio_endio against our io_bio so that we get proper resource
8519 * cleanup if we fail to submit the dip, otherwise, we must do the
8520 * same as btrfs_endio_direct_[write|read] because we can't call these
8521 * callbacks - they require an allocated dip and a clone of dio_bio.
8523 if (io_bio
&& dip
) {
8524 io_bio
->bi_error
= -EIO
;
8527 * The end io callbacks free our dip, do the final put on io_bio
8528 * and all the cleanup and final put for dio_bio (through
8535 btrfs_endio_direct_write_update_ordered(inode
,
8537 dio_bio
->bi_iter
.bi_size
,
8540 unlock_extent(&BTRFS_I(inode
)->io_tree
, file_offset
,
8541 file_offset
+ dio_bio
->bi_iter
.bi_size
- 1);
8543 dio_bio
->bi_error
= -EIO
;
8545 * Releases and cleans up our dio_bio, no need to bio_put()
8546 * nor bio_endio()/bio_io_error() against dio_bio.
8548 dio_end_io(dio_bio
, ret
);
8555 static ssize_t
check_direct_IO(struct btrfs_root
*root
, struct kiocb
*iocb
,
8556 const struct iov_iter
*iter
, loff_t offset
)
8560 unsigned blocksize_mask
= root
->sectorsize
- 1;
8561 ssize_t retval
= -EINVAL
;
8563 if (offset
& blocksize_mask
)
8566 if (iov_iter_alignment(iter
) & blocksize_mask
)
8569 /* If this is a write we don't need to check anymore */
8570 if (iov_iter_rw(iter
) == WRITE
)
8573 * Check to make sure we don't have duplicate iov_base's in this
8574 * iovec, if so return EINVAL, otherwise we'll get csum errors
8575 * when reading back.
8577 for (seg
= 0; seg
< iter
->nr_segs
; seg
++) {
8578 for (i
= seg
+ 1; i
< iter
->nr_segs
; i
++) {
8579 if (iter
->iov
[seg
].iov_base
== iter
->iov
[i
].iov_base
)
8588 static ssize_t
btrfs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
8590 struct file
*file
= iocb
->ki_filp
;
8591 struct inode
*inode
= file
->f_mapping
->host
;
8592 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8593 struct btrfs_dio_data dio_data
= { 0 };
8594 loff_t offset
= iocb
->ki_pos
;
8598 bool relock
= false;
8601 if (check_direct_IO(BTRFS_I(inode
)->root
, iocb
, iter
, offset
))
8604 inode_dio_begin(inode
);
8605 smp_mb__after_atomic();
8608 * The generic stuff only does filemap_write_and_wait_range, which
8609 * isn't enough if we've written compressed pages to this area, so
8610 * we need to flush the dirty pages again to make absolutely sure
8611 * that any outstanding dirty pages are on disk.
8613 count
= iov_iter_count(iter
);
8614 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
8615 &BTRFS_I(inode
)->runtime_flags
))
8616 filemap_fdatawrite_range(inode
->i_mapping
, offset
,
8617 offset
+ count
- 1);
8619 if (iov_iter_rw(iter
) == WRITE
) {
8621 * If the write DIO is beyond the EOF, we need update
8622 * the isize, but it is protected by i_mutex. So we can
8623 * not unlock the i_mutex at this case.
8625 if (offset
+ count
<= inode
->i_size
) {
8626 inode_unlock(inode
);
8629 ret
= btrfs_delalloc_reserve_space(inode
, offset
, count
);
8632 dio_data
.outstanding_extents
= div64_u64(count
+
8633 BTRFS_MAX_EXTENT_SIZE
- 1,
8634 BTRFS_MAX_EXTENT_SIZE
);
8637 * We need to know how many extents we reserved so that we can
8638 * do the accounting properly if we go over the number we
8639 * originally calculated. Abuse current->journal_info for this.
8641 dio_data
.reserve
= round_up(count
, root
->sectorsize
);
8642 dio_data
.unsubmitted_oe_range_start
= (u64
)offset
;
8643 dio_data
.unsubmitted_oe_range_end
= (u64
)offset
;
8644 current
->journal_info
= &dio_data
;
8645 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK
,
8646 &BTRFS_I(inode
)->runtime_flags
)) {
8647 inode_dio_end(inode
);
8648 flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
8652 ret
= __blockdev_direct_IO(iocb
, inode
,
8653 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
8654 iter
, btrfs_get_blocks_direct
, NULL
,
8655 btrfs_submit_direct
, flags
);
8656 if (iov_iter_rw(iter
) == WRITE
) {
8657 current
->journal_info
= NULL
;
8658 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
8659 if (dio_data
.reserve
)
8660 btrfs_delalloc_release_space(inode
, offset
,
8663 * On error we might have left some ordered extents
8664 * without submitting corresponding bios for them, so
8665 * cleanup them up to avoid other tasks getting them
8666 * and waiting for them to complete forever.
8668 if (dio_data
.unsubmitted_oe_range_start
<
8669 dio_data
.unsubmitted_oe_range_end
)
8670 btrfs_endio_direct_write_update_ordered(inode
,
8671 dio_data
.unsubmitted_oe_range_start
,
8672 dio_data
.unsubmitted_oe_range_end
-
8673 dio_data
.unsubmitted_oe_range_start
,
8675 } else if (ret
>= 0 && (size_t)ret
< count
)
8676 btrfs_delalloc_release_space(inode
, offset
,
8677 count
- (size_t)ret
);
8681 inode_dio_end(inode
);
8688 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8690 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
8691 __u64 start
, __u64 len
)
8695 ret
= fiemap_check_flags(fieinfo
, BTRFS_FIEMAP_FLAGS
);
8699 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
8702 int btrfs_readpage(struct file
*file
, struct page
*page
)
8704 struct extent_io_tree
*tree
;
8705 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8706 return extent_read_full_page(tree
, page
, btrfs_get_extent
, 0);
8709 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
8711 struct extent_io_tree
*tree
;
8712 struct inode
*inode
= page
->mapping
->host
;
8715 if (current
->flags
& PF_MEMALLOC
) {
8716 redirty_page_for_writepage(wbc
, page
);
8722 * If we are under memory pressure we will call this directly from the
8723 * VM, we need to make sure we have the inode referenced for the ordered
8724 * extent. If not just return like we didn't do anything.
8726 if (!igrab(inode
)) {
8727 redirty_page_for_writepage(wbc
, page
);
8728 return AOP_WRITEPAGE_ACTIVATE
;
8730 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8731 ret
= extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
8732 btrfs_add_delayed_iput(inode
);
8736 static int btrfs_writepages(struct address_space
*mapping
,
8737 struct writeback_control
*wbc
)
8739 struct extent_io_tree
*tree
;
8741 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
8742 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
8746 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
8747 struct list_head
*pages
, unsigned nr_pages
)
8749 struct extent_io_tree
*tree
;
8750 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
8751 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
8754 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8756 struct extent_io_tree
*tree
;
8757 struct extent_map_tree
*map
;
8760 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8761 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
8762 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
8764 ClearPagePrivate(page
);
8765 set_page_private(page
, 0);
8771 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8773 if (PageWriteback(page
) || PageDirty(page
))
8775 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
8778 static void btrfs_invalidatepage(struct page
*page
, unsigned int offset
,
8779 unsigned int length
)
8781 struct inode
*inode
= page
->mapping
->host
;
8782 struct extent_io_tree
*tree
;
8783 struct btrfs_ordered_extent
*ordered
;
8784 struct extent_state
*cached_state
= NULL
;
8785 u64 page_start
= page_offset(page
);
8786 u64 page_end
= page_start
+ PAGE_SIZE
- 1;
8789 int inode_evicting
= inode
->i_state
& I_FREEING
;
8792 * we have the page locked, so new writeback can't start,
8793 * and the dirty bit won't be cleared while we are here.
8795 * Wait for IO on this page so that we can safely clear
8796 * the PagePrivate2 bit and do ordered accounting
8798 wait_on_page_writeback(page
);
8800 tree
= &BTRFS_I(inode
)->io_tree
;
8802 btrfs_releasepage(page
, GFP_NOFS
);
8806 if (!inode_evicting
)
8807 lock_extent_bits(tree
, page_start
, page_end
, &cached_state
);
8810 ordered
= btrfs_lookup_ordered_range(inode
, start
,
8811 page_end
- start
+ 1);
8813 end
= min(page_end
, ordered
->file_offset
+ ordered
->len
- 1);
8815 * IO on this page will never be started, so we need
8816 * to account for any ordered extents now
8818 if (!inode_evicting
)
8819 clear_extent_bit(tree
, start
, end
,
8820 EXTENT_DIRTY
| EXTENT_DELALLOC
|
8821 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
8822 EXTENT_DEFRAG
, 1, 0, &cached_state
,
8825 * whoever cleared the private bit is responsible
8826 * for the finish_ordered_io
8828 if (TestClearPagePrivate2(page
)) {
8829 struct btrfs_ordered_inode_tree
*tree
;
8832 tree
= &BTRFS_I(inode
)->ordered_tree
;
8834 spin_lock_irq(&tree
->lock
);
8835 set_bit(BTRFS_ORDERED_TRUNCATED
, &ordered
->flags
);
8836 new_len
= start
- ordered
->file_offset
;
8837 if (new_len
< ordered
->truncated_len
)
8838 ordered
->truncated_len
= new_len
;
8839 spin_unlock_irq(&tree
->lock
);
8841 if (btrfs_dec_test_ordered_pending(inode
, &ordered
,
8843 end
- start
+ 1, 1))
8844 btrfs_finish_ordered_io(ordered
);
8846 btrfs_put_ordered_extent(ordered
);
8847 if (!inode_evicting
) {
8848 cached_state
= NULL
;
8849 lock_extent_bits(tree
, start
, end
,
8854 if (start
< page_end
)
8859 * Qgroup reserved space handler
8860 * Page here will be either
8861 * 1) Already written to disk
8862 * In this case, its reserved space is released from data rsv map
8863 * and will be freed by delayed_ref handler finally.
8864 * So even we call qgroup_free_data(), it won't decrease reserved
8866 * 2) Not written to disk
8867 * This means the reserved space should be freed here.
8869 btrfs_qgroup_free_data(inode
, page_start
, PAGE_SIZE
);
8870 if (!inode_evicting
) {
8871 clear_extent_bit(tree
, page_start
, page_end
,
8872 EXTENT_LOCKED
| EXTENT_DIRTY
|
8873 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
8874 EXTENT_DEFRAG
, 1, 1,
8875 &cached_state
, GFP_NOFS
);
8877 __btrfs_releasepage(page
, GFP_NOFS
);
8880 ClearPageChecked(page
);
8881 if (PagePrivate(page
)) {
8882 ClearPagePrivate(page
);
8883 set_page_private(page
, 0);
8889 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8890 * called from a page fault handler when a page is first dirtied. Hence we must
8891 * be careful to check for EOF conditions here. We set the page up correctly
8892 * for a written page which means we get ENOSPC checking when writing into
8893 * holes and correct delalloc and unwritten extent mapping on filesystems that
8894 * support these features.
8896 * We are not allowed to take the i_mutex here so we have to play games to
8897 * protect against truncate races as the page could now be beyond EOF. Because
8898 * vmtruncate() writes the inode size before removing pages, once we have the
8899 * page lock we can determine safely if the page is beyond EOF. If it is not
8900 * beyond EOF, then the page is guaranteed safe against truncation until we
8903 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
8905 struct page
*page
= vmf
->page
;
8906 struct inode
*inode
= file_inode(vma
->vm_file
);
8907 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8908 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
8909 struct btrfs_ordered_extent
*ordered
;
8910 struct extent_state
*cached_state
= NULL
;
8912 unsigned long zero_start
;
8921 reserved_space
= PAGE_SIZE
;
8923 sb_start_pagefault(inode
->i_sb
);
8924 page_start
= page_offset(page
);
8925 page_end
= page_start
+ PAGE_SIZE
- 1;
8929 * Reserving delalloc space after obtaining the page lock can lead to
8930 * deadlock. For example, if a dirty page is locked by this function
8931 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8932 * dirty page write out, then the btrfs_writepage() function could
8933 * end up waiting indefinitely to get a lock on the page currently
8934 * being processed by btrfs_page_mkwrite() function.
8936 ret
= btrfs_delalloc_reserve_space(inode
, page_start
,
8939 ret
= file_update_time(vma
->vm_file
);
8945 else /* -ENOSPC, -EIO, etc */
8946 ret
= VM_FAULT_SIGBUS
;
8952 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
8955 size
= i_size_read(inode
);
8957 if ((page
->mapping
!= inode
->i_mapping
) ||
8958 (page_start
>= size
)) {
8959 /* page got truncated out from underneath us */
8962 wait_on_page_writeback(page
);
8964 lock_extent_bits(io_tree
, page_start
, page_end
, &cached_state
);
8965 set_page_extent_mapped(page
);
8968 * we can't set the delalloc bits if there are pending ordered
8969 * extents. Drop our locks and wait for them to finish
8971 ordered
= btrfs_lookup_ordered_range(inode
, page_start
, page_end
);
8973 unlock_extent_cached(io_tree
, page_start
, page_end
,
8974 &cached_state
, GFP_NOFS
);
8976 btrfs_start_ordered_extent(inode
, ordered
, 1);
8977 btrfs_put_ordered_extent(ordered
);
8981 if (page
->index
== ((size
- 1) >> PAGE_SHIFT
)) {
8982 reserved_space
= round_up(size
- page_start
, root
->sectorsize
);
8983 if (reserved_space
< PAGE_SIZE
) {
8984 end
= page_start
+ reserved_space
- 1;
8985 spin_lock(&BTRFS_I(inode
)->lock
);
8986 BTRFS_I(inode
)->outstanding_extents
++;
8987 spin_unlock(&BTRFS_I(inode
)->lock
);
8988 btrfs_delalloc_release_space(inode
, page_start
,
8989 PAGE_SIZE
- reserved_space
);
8994 * XXX - page_mkwrite gets called every time the page is dirtied, even
8995 * if it was already dirty, so for space accounting reasons we need to
8996 * clear any delalloc bits for the range we are fixing to save. There
8997 * is probably a better way to do this, but for now keep consistent with
8998 * prepare_pages in the normal write path.
9000 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, end
,
9001 EXTENT_DIRTY
| EXTENT_DELALLOC
|
9002 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
9003 0, 0, &cached_state
, GFP_NOFS
);
9005 ret
= btrfs_set_extent_delalloc(inode
, page_start
, end
,
9008 unlock_extent_cached(io_tree
, page_start
, page_end
,
9009 &cached_state
, GFP_NOFS
);
9010 ret
= VM_FAULT_SIGBUS
;
9015 /* page is wholly or partially inside EOF */
9016 if (page_start
+ PAGE_SIZE
> size
)
9017 zero_start
= size
& ~PAGE_MASK
;
9019 zero_start
= PAGE_SIZE
;
9021 if (zero_start
!= PAGE_SIZE
) {
9023 memset(kaddr
+ zero_start
, 0, PAGE_SIZE
- zero_start
);
9024 flush_dcache_page(page
);
9027 ClearPageChecked(page
);
9028 set_page_dirty(page
);
9029 SetPageUptodate(page
);
9031 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
9032 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
9033 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->root
->last_log_commit
;
9035 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
9039 sb_end_pagefault(inode
->i_sb
);
9040 return VM_FAULT_LOCKED
;
9044 btrfs_delalloc_release_space(inode
, page_start
, reserved_space
);
9046 sb_end_pagefault(inode
->i_sb
);
9050 static int btrfs_truncate(struct inode
*inode
)
9052 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9053 struct btrfs_block_rsv
*rsv
;
9056 struct btrfs_trans_handle
*trans
;
9057 u64 mask
= root
->sectorsize
- 1;
9058 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
9060 ret
= btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
),
9066 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9067 * 3 things going on here
9069 * 1) We need to reserve space for our orphan item and the space to
9070 * delete our orphan item. Lord knows we don't want to have a dangling
9071 * orphan item because we didn't reserve space to remove it.
9073 * 2) We need to reserve space to update our inode.
9075 * 3) We need to have something to cache all the space that is going to
9076 * be free'd up by the truncate operation, but also have some slack
9077 * space reserved in case it uses space during the truncate (thank you
9078 * very much snapshotting).
9080 * And we need these to all be separate. The fact is we can use a lot of
9081 * space doing the truncate, and we have no earthly idea how much space
9082 * we will use, so we need the truncate reservation to be separate so it
9083 * doesn't end up using space reserved for updating the inode or
9084 * removing the orphan item. We also need to be able to stop the
9085 * transaction and start a new one, which means we need to be able to
9086 * update the inode several times, and we have no idea of knowing how
9087 * many times that will be, so we can't just reserve 1 item for the
9088 * entirety of the operation, so that has to be done separately as well.
9089 * Then there is the orphan item, which does indeed need to be held on
9090 * to for the whole operation, and we need nobody to touch this reserved
9091 * space except the orphan code.
9093 * So that leaves us with
9095 * 1) root->orphan_block_rsv - for the orphan deletion.
9096 * 2) rsv - for the truncate reservation, which we will steal from the
9097 * transaction reservation.
9098 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9099 * updating the inode.
9101 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
9104 rsv
->size
= min_size
;
9108 * 1 for the truncate slack space
9109 * 1 for updating the inode.
9111 trans
= btrfs_start_transaction(root
, 2);
9112 if (IS_ERR(trans
)) {
9113 err
= PTR_ERR(trans
);
9117 /* Migrate the slack space for the truncate to our reserve */
9118 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
, rsv
,
9123 * So if we truncate and then write and fsync we normally would just
9124 * write the extents that changed, which is a problem if we need to
9125 * first truncate that entire inode. So set this flag so we write out
9126 * all of the extents in the inode to the sync log so we're completely
9129 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
9130 trans
->block_rsv
= rsv
;
9133 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
9135 BTRFS_EXTENT_DATA_KEY
);
9136 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
) {
9141 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
9142 ret
= btrfs_update_inode(trans
, root
, inode
);
9148 btrfs_end_transaction(trans
, root
);
9149 btrfs_btree_balance_dirty(root
);
9151 trans
= btrfs_start_transaction(root
, 2);
9152 if (IS_ERR(trans
)) {
9153 ret
= err
= PTR_ERR(trans
);
9158 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
,
9160 BUG_ON(ret
); /* shouldn't happen */
9161 trans
->block_rsv
= rsv
;
9164 if (ret
== 0 && inode
->i_nlink
> 0) {
9165 trans
->block_rsv
= root
->orphan_block_rsv
;
9166 ret
= btrfs_orphan_del(trans
, inode
);
9172 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
9173 ret
= btrfs_update_inode(trans
, root
, inode
);
9177 ret
= btrfs_end_transaction(trans
, root
);
9178 btrfs_btree_balance_dirty(root
);
9182 btrfs_free_block_rsv(root
, rsv
);
9191 * create a new subvolume directory/inode (helper for the ioctl).
9193 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
9194 struct btrfs_root
*new_root
,
9195 struct btrfs_root
*parent_root
,
9198 struct inode
*inode
;
9202 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
9203 new_dirid
, new_dirid
,
9204 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
9207 return PTR_ERR(inode
);
9208 inode
->i_op
= &btrfs_dir_inode_operations
;
9209 inode
->i_fop
= &btrfs_dir_file_operations
;
9211 set_nlink(inode
, 1);
9212 btrfs_i_size_write(inode
, 0);
9213 unlock_new_inode(inode
);
9215 err
= btrfs_subvol_inherit_props(trans
, new_root
, parent_root
);
9217 btrfs_err(new_root
->fs_info
,
9218 "error inheriting subvolume %llu properties: %d",
9219 new_root
->root_key
.objectid
, err
);
9221 err
= btrfs_update_inode(trans
, new_root
, inode
);
9227 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
9229 struct btrfs_inode
*ei
;
9230 struct inode
*inode
;
9232 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
9239 ei
->last_sub_trans
= 0;
9240 ei
->logged_trans
= 0;
9241 ei
->delalloc_bytes
= 0;
9242 ei
->defrag_bytes
= 0;
9243 ei
->disk_i_size
= 0;
9246 ei
->index_cnt
= (u64
)-1;
9248 ei
->last_unlink_trans
= 0;
9249 ei
->last_log_commit
= 0;
9250 ei
->delayed_iput_count
= 0;
9252 spin_lock_init(&ei
->lock
);
9253 ei
->outstanding_extents
= 0;
9254 ei
->reserved_extents
= 0;
9256 ei
->runtime_flags
= 0;
9257 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
9259 ei
->delayed_node
= NULL
;
9261 ei
->i_otime
.tv_sec
= 0;
9262 ei
->i_otime
.tv_nsec
= 0;
9264 inode
= &ei
->vfs_inode
;
9265 extent_map_tree_init(&ei
->extent_tree
);
9266 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
);
9267 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
);
9268 ei
->io_tree
.track_uptodate
= 1;
9269 ei
->io_failure_tree
.track_uptodate
= 1;
9270 atomic_set(&ei
->sync_writers
, 0);
9271 mutex_init(&ei
->log_mutex
);
9272 mutex_init(&ei
->delalloc_mutex
);
9273 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
9274 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
9275 INIT_LIST_HEAD(&ei
->delayed_iput
);
9276 RB_CLEAR_NODE(&ei
->rb_node
);
9277 init_rwsem(&ei
->dio_sem
);
9282 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9283 void btrfs_test_destroy_inode(struct inode
*inode
)
9285 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
9286 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9290 static void btrfs_i_callback(struct rcu_head
*head
)
9292 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
9293 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9296 void btrfs_destroy_inode(struct inode
*inode
)
9298 struct btrfs_ordered_extent
*ordered
;
9299 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9301 WARN_ON(!hlist_empty(&inode
->i_dentry
));
9302 WARN_ON(inode
->i_data
.nrpages
);
9303 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
9304 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
9305 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
9306 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
9307 WARN_ON(BTRFS_I(inode
)->defrag_bytes
);
9310 * This can happen where we create an inode, but somebody else also
9311 * created the same inode and we need to destroy the one we already
9317 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
9318 &BTRFS_I(inode
)->runtime_flags
)) {
9319 btrfs_info(root
->fs_info
, "inode %llu still on the orphan list",
9321 atomic_dec(&root
->orphan_inodes
);
9325 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
9329 btrfs_err(root
->fs_info
, "found ordered extent %llu %llu on inode cleanup",
9330 ordered
->file_offset
, ordered
->len
);
9331 btrfs_remove_ordered_extent(inode
, ordered
);
9332 btrfs_put_ordered_extent(ordered
);
9333 btrfs_put_ordered_extent(ordered
);
9336 btrfs_qgroup_check_reserved_leak(inode
);
9337 inode_tree_del(inode
);
9338 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
9340 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
9343 int btrfs_drop_inode(struct inode
*inode
)
9345 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9350 /* the snap/subvol tree is on deleting */
9351 if (btrfs_root_refs(&root
->root_item
) == 0)
9354 return generic_drop_inode(inode
);
9357 static void init_once(void *foo
)
9359 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
9361 inode_init_once(&ei
->vfs_inode
);
9364 void btrfs_destroy_cachep(void)
9367 * Make sure all delayed rcu free inodes are flushed before we
9371 kmem_cache_destroy(btrfs_inode_cachep
);
9372 kmem_cache_destroy(btrfs_trans_handle_cachep
);
9373 kmem_cache_destroy(btrfs_transaction_cachep
);
9374 kmem_cache_destroy(btrfs_path_cachep
);
9375 kmem_cache_destroy(btrfs_free_space_cachep
);
9378 int btrfs_init_cachep(void)
9380 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode",
9381 sizeof(struct btrfs_inode
), 0,
9382 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
| SLAB_ACCOUNT
,
9384 if (!btrfs_inode_cachep
)
9387 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle",
9388 sizeof(struct btrfs_trans_handle
), 0,
9389 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9390 if (!btrfs_trans_handle_cachep
)
9393 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction",
9394 sizeof(struct btrfs_transaction
), 0,
9395 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9396 if (!btrfs_transaction_cachep
)
9399 btrfs_path_cachep
= kmem_cache_create("btrfs_path",
9400 sizeof(struct btrfs_path
), 0,
9401 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9402 if (!btrfs_path_cachep
)
9405 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space",
9406 sizeof(struct btrfs_free_space
), 0,
9407 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9408 if (!btrfs_free_space_cachep
)
9413 btrfs_destroy_cachep();
9417 static int btrfs_getattr(struct vfsmount
*mnt
,
9418 struct dentry
*dentry
, struct kstat
*stat
)
9421 struct inode
*inode
= d_inode(dentry
);
9422 u32 blocksize
= inode
->i_sb
->s_blocksize
;
9424 generic_fillattr(inode
, stat
);
9425 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
9427 spin_lock(&BTRFS_I(inode
)->lock
);
9428 delalloc_bytes
= BTRFS_I(inode
)->delalloc_bytes
;
9429 spin_unlock(&BTRFS_I(inode
)->lock
);
9430 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
9431 ALIGN(delalloc_bytes
, blocksize
)) >> 9;
9435 static int btrfs_rename_exchange(struct inode
*old_dir
,
9436 struct dentry
*old_dentry
,
9437 struct inode
*new_dir
,
9438 struct dentry
*new_dentry
)
9440 struct btrfs_trans_handle
*trans
;
9441 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
9442 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
9443 struct inode
*new_inode
= new_dentry
->d_inode
;
9444 struct inode
*old_inode
= old_dentry
->d_inode
;
9445 struct timespec ctime
= CURRENT_TIME
;
9446 struct dentry
*parent
;
9447 u64 old_ino
= btrfs_ino(old_inode
);
9448 u64 new_ino
= btrfs_ino(new_inode
);
9453 bool root_log_pinned
= false;
9454 bool dest_log_pinned
= false;
9456 /* we only allow rename subvolume link between subvolumes */
9457 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
9460 /* close the race window with snapshot create/destroy ioctl */
9461 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9462 down_read(&root
->fs_info
->subvol_sem
);
9463 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9464 down_read(&dest
->fs_info
->subvol_sem
);
9467 * We want to reserve the absolute worst case amount of items. So if
9468 * both inodes are subvols and we need to unlink them then that would
9469 * require 4 item modifications, but if they are both normal inodes it
9470 * would require 5 item modifications, so we'll assume their normal
9471 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9472 * should cover the worst case number of items we'll modify.
9474 trans
= btrfs_start_transaction(root
, 12);
9475 if (IS_ERR(trans
)) {
9476 ret
= PTR_ERR(trans
);
9481 * We need to find a free sequence number both in the source and
9482 * in the destination directory for the exchange.
9484 ret
= btrfs_set_inode_index(new_dir
, &old_idx
);
9487 ret
= btrfs_set_inode_index(old_dir
, &new_idx
);
9491 BTRFS_I(old_inode
)->dir_index
= 0ULL;
9492 BTRFS_I(new_inode
)->dir_index
= 0ULL;
9494 /* Reference for the source. */
9495 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9496 /* force full log commit if subvolume involved. */
9497 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9499 btrfs_pin_log_trans(root
);
9500 root_log_pinned
= true;
9501 ret
= btrfs_insert_inode_ref(trans
, dest
,
9502 new_dentry
->d_name
.name
,
9503 new_dentry
->d_name
.len
,
9505 btrfs_ino(new_dir
), old_idx
);
9510 /* And now for the dest. */
9511 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9512 /* force full log commit if subvolume involved. */
9513 btrfs_set_log_full_commit(dest
->fs_info
, trans
);
9515 btrfs_pin_log_trans(dest
);
9516 dest_log_pinned
= true;
9517 ret
= btrfs_insert_inode_ref(trans
, root
,
9518 old_dentry
->d_name
.name
,
9519 old_dentry
->d_name
.len
,
9521 btrfs_ino(old_dir
), new_idx
);
9526 /* Update inode version and ctime/mtime. */
9527 inode_inc_iversion(old_dir
);
9528 inode_inc_iversion(new_dir
);
9529 inode_inc_iversion(old_inode
);
9530 inode_inc_iversion(new_inode
);
9531 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
9532 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
9533 old_inode
->i_ctime
= ctime
;
9534 new_inode
->i_ctime
= ctime
;
9536 if (old_dentry
->d_parent
!= new_dentry
->d_parent
) {
9537 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
9538 btrfs_record_unlink_dir(trans
, new_dir
, new_inode
, 1);
9541 /* src is a subvolume */
9542 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9543 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
9544 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
,
9546 old_dentry
->d_name
.name
,
9547 old_dentry
->d_name
.len
);
9548 } else { /* src is an inode */
9549 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
9550 old_dentry
->d_inode
,
9551 old_dentry
->d_name
.name
,
9552 old_dentry
->d_name
.len
);
9554 ret
= btrfs_update_inode(trans
, root
, old_inode
);
9557 btrfs_abort_transaction(trans
, root
, ret
);
9561 /* dest is a subvolume */
9562 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9563 root_objectid
= BTRFS_I(new_inode
)->root
->root_key
.objectid
;
9564 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
9566 new_dentry
->d_name
.name
,
9567 new_dentry
->d_name
.len
);
9568 } else { /* dest is an inode */
9569 ret
= __btrfs_unlink_inode(trans
, dest
, new_dir
,
9570 new_dentry
->d_inode
,
9571 new_dentry
->d_name
.name
,
9572 new_dentry
->d_name
.len
);
9574 ret
= btrfs_update_inode(trans
, dest
, new_inode
);
9577 btrfs_abort_transaction(trans
, root
, ret
);
9581 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
9582 new_dentry
->d_name
.name
,
9583 new_dentry
->d_name
.len
, 0, old_idx
);
9585 btrfs_abort_transaction(trans
, root
, ret
);
9589 ret
= btrfs_add_link(trans
, old_dir
, new_inode
,
9590 old_dentry
->d_name
.name
,
9591 old_dentry
->d_name
.len
, 0, new_idx
);
9593 btrfs_abort_transaction(trans
, root
, ret
);
9597 if (old_inode
->i_nlink
== 1)
9598 BTRFS_I(old_inode
)->dir_index
= old_idx
;
9599 if (new_inode
->i_nlink
== 1)
9600 BTRFS_I(new_inode
)->dir_index
= new_idx
;
9602 if (root_log_pinned
) {
9603 parent
= new_dentry
->d_parent
;
9604 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
9605 btrfs_end_log_trans(root
);
9606 root_log_pinned
= false;
9608 if (dest_log_pinned
) {
9609 parent
= old_dentry
->d_parent
;
9610 btrfs_log_new_name(trans
, new_inode
, new_dir
, parent
);
9611 btrfs_end_log_trans(dest
);
9612 dest_log_pinned
= false;
9616 * If we have pinned a log and an error happened, we unpin tasks
9617 * trying to sync the log and force them to fallback to a transaction
9618 * commit if the log currently contains any of the inodes involved in
9619 * this rename operation (to ensure we do not persist a log with an
9620 * inconsistent state for any of these inodes or leading to any
9621 * inconsistencies when replayed). If the transaction was aborted, the
9622 * abortion reason is propagated to userspace when attempting to commit
9623 * the transaction. If the log does not contain any of these inodes, we
9624 * allow the tasks to sync it.
9626 if (ret
&& (root_log_pinned
|| dest_log_pinned
)) {
9627 if (btrfs_inode_in_log(old_dir
, root
->fs_info
->generation
) ||
9628 btrfs_inode_in_log(new_dir
, root
->fs_info
->generation
) ||
9629 btrfs_inode_in_log(old_inode
, root
->fs_info
->generation
) ||
9631 btrfs_inode_in_log(new_inode
, root
->fs_info
->generation
)))
9632 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9634 if (root_log_pinned
) {
9635 btrfs_end_log_trans(root
);
9636 root_log_pinned
= false;
9638 if (dest_log_pinned
) {
9639 btrfs_end_log_trans(dest
);
9640 dest_log_pinned
= false;
9643 ret
= btrfs_end_transaction(trans
, root
);
9645 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9646 up_read(&dest
->fs_info
->subvol_sem
);
9647 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9648 up_read(&root
->fs_info
->subvol_sem
);
9653 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle
*trans
,
9654 struct btrfs_root
*root
,
9656 struct dentry
*dentry
)
9659 struct inode
*inode
;
9663 ret
= btrfs_find_free_ino(root
, &objectid
);
9667 inode
= btrfs_new_inode(trans
, root
, dir
,
9668 dentry
->d_name
.name
,
9672 S_IFCHR
| WHITEOUT_MODE
,
9675 if (IS_ERR(inode
)) {
9676 ret
= PTR_ERR(inode
);
9680 inode
->i_op
= &btrfs_special_inode_operations
;
9681 init_special_inode(inode
, inode
->i_mode
,
9684 ret
= btrfs_init_inode_security(trans
, inode
, dir
,
9689 ret
= btrfs_add_nondir(trans
, dir
, dentry
,
9694 ret
= btrfs_update_inode(trans
, root
, inode
);
9696 unlock_new_inode(inode
);
9698 inode_dec_link_count(inode
);
9704 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
9705 struct inode
*new_dir
, struct dentry
*new_dentry
,
9708 struct btrfs_trans_handle
*trans
;
9709 unsigned int trans_num_items
;
9710 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
9711 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
9712 struct inode
*new_inode
= d_inode(new_dentry
);
9713 struct inode
*old_inode
= d_inode(old_dentry
);
9717 u64 old_ino
= btrfs_ino(old_inode
);
9718 bool log_pinned
= false;
9720 if (btrfs_ino(new_dir
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
9723 /* we only allow rename subvolume link between subvolumes */
9724 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
9727 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
9728 (new_inode
&& btrfs_ino(new_inode
) == BTRFS_FIRST_FREE_OBJECTID
))
9731 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
9732 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
9736 /* check for collisions, even if the name isn't there */
9737 ret
= btrfs_check_dir_item_collision(dest
, new_dir
->i_ino
,
9738 new_dentry
->d_name
.name
,
9739 new_dentry
->d_name
.len
);
9742 if (ret
== -EEXIST
) {
9744 * eexist without a new_inode */
9745 if (WARN_ON(!new_inode
)) {
9749 /* maybe -EOVERFLOW */
9756 * we're using rename to replace one file with another. Start IO on it
9757 * now so we don't add too much work to the end of the transaction
9759 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
)
9760 filemap_flush(old_inode
->i_mapping
);
9762 /* close the racy window with snapshot create/destroy ioctl */
9763 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9764 down_read(&root
->fs_info
->subvol_sem
);
9766 * We want to reserve the absolute worst case amount of items. So if
9767 * both inodes are subvols and we need to unlink them then that would
9768 * require 4 item modifications, but if they are both normal inodes it
9769 * would require 5 item modifications, so we'll assume they are normal
9770 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9771 * should cover the worst case number of items we'll modify.
9772 * If our rename has the whiteout flag, we need more 5 units for the
9773 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9774 * when selinux is enabled).
9776 trans_num_items
= 11;
9777 if (flags
& RENAME_WHITEOUT
)
9778 trans_num_items
+= 5;
9779 trans
= btrfs_start_transaction(root
, trans_num_items
);
9780 if (IS_ERR(trans
)) {
9781 ret
= PTR_ERR(trans
);
9786 btrfs_record_root_in_trans(trans
, dest
);
9788 ret
= btrfs_set_inode_index(new_dir
, &index
);
9792 BTRFS_I(old_inode
)->dir_index
= 0ULL;
9793 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9794 /* force full log commit if subvolume involved. */
9795 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9797 btrfs_pin_log_trans(root
);
9799 ret
= btrfs_insert_inode_ref(trans
, dest
,
9800 new_dentry
->d_name
.name
,
9801 new_dentry
->d_name
.len
,
9803 btrfs_ino(new_dir
), index
);
9808 inode_inc_iversion(old_dir
);
9809 inode_inc_iversion(new_dir
);
9810 inode_inc_iversion(old_inode
);
9811 old_dir
->i_ctime
= old_dir
->i_mtime
=
9812 new_dir
->i_ctime
= new_dir
->i_mtime
=
9813 old_inode
->i_ctime
= current_fs_time(old_dir
->i_sb
);
9815 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
9816 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
9818 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9819 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
9820 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
9821 old_dentry
->d_name
.name
,
9822 old_dentry
->d_name
.len
);
9824 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
9825 d_inode(old_dentry
),
9826 old_dentry
->d_name
.name
,
9827 old_dentry
->d_name
.len
);
9829 ret
= btrfs_update_inode(trans
, root
, old_inode
);
9832 btrfs_abort_transaction(trans
, root
, ret
);
9837 inode_inc_iversion(new_inode
);
9838 new_inode
->i_ctime
= current_fs_time(new_inode
->i_sb
);
9839 if (unlikely(btrfs_ino(new_inode
) ==
9840 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
9841 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
9842 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
9844 new_dentry
->d_name
.name
,
9845 new_dentry
->d_name
.len
);
9846 BUG_ON(new_inode
->i_nlink
== 0);
9848 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
9849 d_inode(new_dentry
),
9850 new_dentry
->d_name
.name
,
9851 new_dentry
->d_name
.len
);
9853 if (!ret
&& new_inode
->i_nlink
== 0)
9854 ret
= btrfs_orphan_add(trans
, d_inode(new_dentry
));
9856 btrfs_abort_transaction(trans
, root
, ret
);
9861 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
9862 new_dentry
->d_name
.name
,
9863 new_dentry
->d_name
.len
, 0, index
);
9865 btrfs_abort_transaction(trans
, root
, ret
);
9869 if (old_inode
->i_nlink
== 1)
9870 BTRFS_I(old_inode
)->dir_index
= index
;
9873 struct dentry
*parent
= new_dentry
->d_parent
;
9875 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
9876 btrfs_end_log_trans(root
);
9880 if (flags
& RENAME_WHITEOUT
) {
9881 ret
= btrfs_whiteout_for_rename(trans
, root
, old_dir
,
9885 btrfs_abort_transaction(trans
, root
, ret
);
9891 * If we have pinned the log and an error happened, we unpin tasks
9892 * trying to sync the log and force them to fallback to a transaction
9893 * commit if the log currently contains any of the inodes involved in
9894 * this rename operation (to ensure we do not persist a log with an
9895 * inconsistent state for any of these inodes or leading to any
9896 * inconsistencies when replayed). If the transaction was aborted, the
9897 * abortion reason is propagated to userspace when attempting to commit
9898 * the transaction. If the log does not contain any of these inodes, we
9899 * allow the tasks to sync it.
9901 if (ret
&& log_pinned
) {
9902 if (btrfs_inode_in_log(old_dir
, root
->fs_info
->generation
) ||
9903 btrfs_inode_in_log(new_dir
, root
->fs_info
->generation
) ||
9904 btrfs_inode_in_log(old_inode
, root
->fs_info
->generation
) ||
9906 btrfs_inode_in_log(new_inode
, root
->fs_info
->generation
)))
9907 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9909 btrfs_end_log_trans(root
);
9912 btrfs_end_transaction(trans
, root
);
9914 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9915 up_read(&root
->fs_info
->subvol_sem
);
9920 static int btrfs_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
9921 struct inode
*new_dir
, struct dentry
*new_dentry
,
9924 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
9927 if (flags
& RENAME_EXCHANGE
)
9928 return btrfs_rename_exchange(old_dir
, old_dentry
, new_dir
,
9931 return btrfs_rename(old_dir
, old_dentry
, new_dir
, new_dentry
, flags
);
9934 static void btrfs_run_delalloc_work(struct btrfs_work
*work
)
9936 struct btrfs_delalloc_work
*delalloc_work
;
9937 struct inode
*inode
;
9939 delalloc_work
= container_of(work
, struct btrfs_delalloc_work
,
9941 inode
= delalloc_work
->inode
;
9942 filemap_flush(inode
->i_mapping
);
9943 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
9944 &BTRFS_I(inode
)->runtime_flags
))
9945 filemap_flush(inode
->i_mapping
);
9947 if (delalloc_work
->delay_iput
)
9948 btrfs_add_delayed_iput(inode
);
9951 complete(&delalloc_work
->completion
);
9954 struct btrfs_delalloc_work
*btrfs_alloc_delalloc_work(struct inode
*inode
,
9957 struct btrfs_delalloc_work
*work
;
9959 work
= kmalloc(sizeof(*work
), GFP_NOFS
);
9963 init_completion(&work
->completion
);
9964 INIT_LIST_HEAD(&work
->list
);
9965 work
->inode
= inode
;
9966 work
->delay_iput
= delay_iput
;
9967 WARN_ON_ONCE(!inode
);
9968 btrfs_init_work(&work
->work
, btrfs_flush_delalloc_helper
,
9969 btrfs_run_delalloc_work
, NULL
, NULL
);
9974 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work
*work
)
9976 wait_for_completion(&work
->completion
);
9981 * some fairly slow code that needs optimization. This walks the list
9982 * of all the inodes with pending delalloc and forces them to disk.
9984 static int __start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
,
9987 struct btrfs_inode
*binode
;
9988 struct inode
*inode
;
9989 struct btrfs_delalloc_work
*work
, *next
;
9990 struct list_head works
;
9991 struct list_head splice
;
9994 INIT_LIST_HEAD(&works
);
9995 INIT_LIST_HEAD(&splice
);
9997 mutex_lock(&root
->delalloc_mutex
);
9998 spin_lock(&root
->delalloc_lock
);
9999 list_splice_init(&root
->delalloc_inodes
, &splice
);
10000 while (!list_empty(&splice
)) {
10001 binode
= list_entry(splice
.next
, struct btrfs_inode
,
10004 list_move_tail(&binode
->delalloc_inodes
,
10005 &root
->delalloc_inodes
);
10006 inode
= igrab(&binode
->vfs_inode
);
10008 cond_resched_lock(&root
->delalloc_lock
);
10011 spin_unlock(&root
->delalloc_lock
);
10013 work
= btrfs_alloc_delalloc_work(inode
, delay_iput
);
10016 btrfs_add_delayed_iput(inode
);
10022 list_add_tail(&work
->list
, &works
);
10023 btrfs_queue_work(root
->fs_info
->flush_workers
,
10026 if (nr
!= -1 && ret
>= nr
)
10029 spin_lock(&root
->delalloc_lock
);
10031 spin_unlock(&root
->delalloc_lock
);
10034 list_for_each_entry_safe(work
, next
, &works
, list
) {
10035 list_del_init(&work
->list
);
10036 btrfs_wait_and_free_delalloc_work(work
);
10039 if (!list_empty_careful(&splice
)) {
10040 spin_lock(&root
->delalloc_lock
);
10041 list_splice_tail(&splice
, &root
->delalloc_inodes
);
10042 spin_unlock(&root
->delalloc_lock
);
10044 mutex_unlock(&root
->delalloc_mutex
);
10048 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
10052 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
10055 ret
= __start_delalloc_inodes(root
, delay_iput
, -1);
10059 * the filemap_flush will queue IO into the worker threads, but
10060 * we have to make sure the IO is actually started and that
10061 * ordered extents get created before we return
10063 atomic_inc(&root
->fs_info
->async_submit_draining
);
10064 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
10065 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
10066 wait_event(root
->fs_info
->async_submit_wait
,
10067 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
10068 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
10070 atomic_dec(&root
->fs_info
->async_submit_draining
);
10074 int btrfs_start_delalloc_roots(struct btrfs_fs_info
*fs_info
, int delay_iput
,
10077 struct btrfs_root
*root
;
10078 struct list_head splice
;
10081 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
10084 INIT_LIST_HEAD(&splice
);
10086 mutex_lock(&fs_info
->delalloc_root_mutex
);
10087 spin_lock(&fs_info
->delalloc_root_lock
);
10088 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
10089 while (!list_empty(&splice
) && nr
) {
10090 root
= list_first_entry(&splice
, struct btrfs_root
,
10092 root
= btrfs_grab_fs_root(root
);
10094 list_move_tail(&root
->delalloc_root
,
10095 &fs_info
->delalloc_roots
);
10096 spin_unlock(&fs_info
->delalloc_root_lock
);
10098 ret
= __start_delalloc_inodes(root
, delay_iput
, nr
);
10099 btrfs_put_fs_root(root
);
10107 spin_lock(&fs_info
->delalloc_root_lock
);
10109 spin_unlock(&fs_info
->delalloc_root_lock
);
10112 atomic_inc(&fs_info
->async_submit_draining
);
10113 while (atomic_read(&fs_info
->nr_async_submits
) ||
10114 atomic_read(&fs_info
->async_delalloc_pages
)) {
10115 wait_event(fs_info
->async_submit_wait
,
10116 (atomic_read(&fs_info
->nr_async_submits
) == 0 &&
10117 atomic_read(&fs_info
->async_delalloc_pages
) == 0));
10119 atomic_dec(&fs_info
->async_submit_draining
);
10121 if (!list_empty_careful(&splice
)) {
10122 spin_lock(&fs_info
->delalloc_root_lock
);
10123 list_splice_tail(&splice
, &fs_info
->delalloc_roots
);
10124 spin_unlock(&fs_info
->delalloc_root_lock
);
10126 mutex_unlock(&fs_info
->delalloc_root_mutex
);
10130 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
10131 const char *symname
)
10133 struct btrfs_trans_handle
*trans
;
10134 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
10135 struct btrfs_path
*path
;
10136 struct btrfs_key key
;
10137 struct inode
*inode
= NULL
;
10139 int drop_inode
= 0;
10145 struct btrfs_file_extent_item
*ei
;
10146 struct extent_buffer
*leaf
;
10148 name_len
= strlen(symname
);
10149 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
10150 return -ENAMETOOLONG
;
10153 * 2 items for inode item and ref
10154 * 2 items for dir items
10155 * 1 item for updating parent inode item
10156 * 1 item for the inline extent item
10157 * 1 item for xattr if selinux is on
10159 trans
= btrfs_start_transaction(root
, 7);
10161 return PTR_ERR(trans
);
10163 err
= btrfs_find_free_ino(root
, &objectid
);
10167 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
10168 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
10169 S_IFLNK
|S_IRWXUGO
, &index
);
10170 if (IS_ERR(inode
)) {
10171 err
= PTR_ERR(inode
);
10176 * If the active LSM wants to access the inode during
10177 * d_instantiate it needs these. Smack checks to see
10178 * if the filesystem supports xattrs by looking at the
10181 inode
->i_fop
= &btrfs_file_operations
;
10182 inode
->i_op
= &btrfs_file_inode_operations
;
10183 inode
->i_mapping
->a_ops
= &btrfs_aops
;
10184 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
10186 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
10188 goto out_unlock_inode
;
10190 path
= btrfs_alloc_path();
10193 goto out_unlock_inode
;
10195 key
.objectid
= btrfs_ino(inode
);
10197 key
.type
= BTRFS_EXTENT_DATA_KEY
;
10198 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
10199 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
10202 btrfs_free_path(path
);
10203 goto out_unlock_inode
;
10205 leaf
= path
->nodes
[0];
10206 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
10207 struct btrfs_file_extent_item
);
10208 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
10209 btrfs_set_file_extent_type(leaf
, ei
,
10210 BTRFS_FILE_EXTENT_INLINE
);
10211 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
10212 btrfs_set_file_extent_compression(leaf
, ei
, 0);
10213 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
10214 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
10216 ptr
= btrfs_file_extent_inline_start(ei
);
10217 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
10218 btrfs_mark_buffer_dirty(leaf
);
10219 btrfs_free_path(path
);
10221 inode
->i_op
= &btrfs_symlink_inode_operations
;
10222 inode_nohighmem(inode
);
10223 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
10224 inode_set_bytes(inode
, name_len
);
10225 btrfs_i_size_write(inode
, name_len
);
10226 err
= btrfs_update_inode(trans
, root
, inode
);
10228 * Last step, add directory indexes for our symlink inode. This is the
10229 * last step to avoid extra cleanup of these indexes if an error happens
10233 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
10236 goto out_unlock_inode
;
10239 unlock_new_inode(inode
);
10240 d_instantiate(dentry
, inode
);
10243 btrfs_end_transaction(trans
, root
);
10245 inode_dec_link_count(inode
);
10248 btrfs_btree_balance_dirty(root
);
10253 unlock_new_inode(inode
);
10257 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
10258 u64 start
, u64 num_bytes
, u64 min_size
,
10259 loff_t actual_len
, u64
*alloc_hint
,
10260 struct btrfs_trans_handle
*trans
)
10262 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
10263 struct extent_map
*em
;
10264 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
10265 struct btrfs_key ins
;
10266 u64 cur_offset
= start
;
10269 u64 last_alloc
= (u64
)-1;
10271 bool own_trans
= true;
10275 while (num_bytes
> 0) {
10277 trans
= btrfs_start_transaction(root
, 3);
10278 if (IS_ERR(trans
)) {
10279 ret
= PTR_ERR(trans
);
10284 cur_bytes
= min_t(u64
, num_bytes
, SZ_256M
);
10285 cur_bytes
= max(cur_bytes
, min_size
);
10287 * If we are severely fragmented we could end up with really
10288 * small allocations, so if the allocator is returning small
10289 * chunks lets make its job easier by only searching for those
10292 cur_bytes
= min(cur_bytes
, last_alloc
);
10293 ret
= btrfs_reserve_extent(root
, cur_bytes
, min_size
, 0,
10294 *alloc_hint
, &ins
, 1, 0);
10297 btrfs_end_transaction(trans
, root
);
10300 btrfs_dec_block_group_reservations(root
->fs_info
, ins
.objectid
);
10302 last_alloc
= ins
.offset
;
10303 ret
= insert_reserved_file_extent(trans
, inode
,
10304 cur_offset
, ins
.objectid
,
10305 ins
.offset
, ins
.offset
,
10306 ins
.offset
, 0, 0, 0,
10307 BTRFS_FILE_EXTENT_PREALLOC
);
10309 btrfs_free_reserved_extent(root
, ins
.objectid
,
10311 btrfs_abort_transaction(trans
, root
, ret
);
10313 btrfs_end_transaction(trans
, root
);
10317 btrfs_drop_extent_cache(inode
, cur_offset
,
10318 cur_offset
+ ins
.offset
-1, 0);
10320 em
= alloc_extent_map();
10322 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
10323 &BTRFS_I(inode
)->runtime_flags
);
10327 em
->start
= cur_offset
;
10328 em
->orig_start
= cur_offset
;
10329 em
->len
= ins
.offset
;
10330 em
->block_start
= ins
.objectid
;
10331 em
->block_len
= ins
.offset
;
10332 em
->orig_block_len
= ins
.offset
;
10333 em
->ram_bytes
= ins
.offset
;
10334 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
10335 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
10336 em
->generation
= trans
->transid
;
10339 write_lock(&em_tree
->lock
);
10340 ret
= add_extent_mapping(em_tree
, em
, 1);
10341 write_unlock(&em_tree
->lock
);
10342 if (ret
!= -EEXIST
)
10344 btrfs_drop_extent_cache(inode
, cur_offset
,
10345 cur_offset
+ ins
.offset
- 1,
10348 free_extent_map(em
);
10350 num_bytes
-= ins
.offset
;
10351 cur_offset
+= ins
.offset
;
10352 *alloc_hint
= ins
.objectid
+ ins
.offset
;
10354 inode_inc_iversion(inode
);
10355 inode
->i_ctime
= current_fs_time(inode
->i_sb
);
10356 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
10357 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
10358 (actual_len
> inode
->i_size
) &&
10359 (cur_offset
> inode
->i_size
)) {
10360 if (cur_offset
> actual_len
)
10361 i_size
= actual_len
;
10363 i_size
= cur_offset
;
10364 i_size_write(inode
, i_size
);
10365 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
10368 ret
= btrfs_update_inode(trans
, root
, inode
);
10371 btrfs_abort_transaction(trans
, root
, ret
);
10373 btrfs_end_transaction(trans
, root
);
10378 btrfs_end_transaction(trans
, root
);
10383 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
10384 u64 start
, u64 num_bytes
, u64 min_size
,
10385 loff_t actual_len
, u64
*alloc_hint
)
10387 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
10388 min_size
, actual_len
, alloc_hint
,
10392 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
10393 struct btrfs_trans_handle
*trans
, int mode
,
10394 u64 start
, u64 num_bytes
, u64 min_size
,
10395 loff_t actual_len
, u64
*alloc_hint
)
10397 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
10398 min_size
, actual_len
, alloc_hint
, trans
);
10401 static int btrfs_set_page_dirty(struct page
*page
)
10403 return __set_page_dirty_nobuffers(page
);
10406 static int btrfs_permission(struct inode
*inode
, int mask
)
10408 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
10409 umode_t mode
= inode
->i_mode
;
10411 if (mask
& MAY_WRITE
&&
10412 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
10413 if (btrfs_root_readonly(root
))
10415 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
10418 return generic_permission(inode
, mask
);
10421 static int btrfs_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
10423 struct btrfs_trans_handle
*trans
;
10424 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
10425 struct inode
*inode
= NULL
;
10431 * 5 units required for adding orphan entry
10433 trans
= btrfs_start_transaction(root
, 5);
10435 return PTR_ERR(trans
);
10437 ret
= btrfs_find_free_ino(root
, &objectid
);
10441 inode
= btrfs_new_inode(trans
, root
, dir
, NULL
, 0,
10442 btrfs_ino(dir
), objectid
, mode
, &index
);
10443 if (IS_ERR(inode
)) {
10444 ret
= PTR_ERR(inode
);
10449 inode
->i_fop
= &btrfs_file_operations
;
10450 inode
->i_op
= &btrfs_file_inode_operations
;
10452 inode
->i_mapping
->a_ops
= &btrfs_aops
;
10453 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
10455 ret
= btrfs_init_inode_security(trans
, inode
, dir
, NULL
);
10459 ret
= btrfs_update_inode(trans
, root
, inode
);
10462 ret
= btrfs_orphan_add(trans
, inode
);
10467 * We set number of links to 0 in btrfs_new_inode(), and here we set
10468 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10471 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10473 set_nlink(inode
, 1);
10474 unlock_new_inode(inode
);
10475 d_tmpfile(dentry
, inode
);
10476 mark_inode_dirty(inode
);
10479 btrfs_end_transaction(trans
, root
);
10482 btrfs_balance_delayed_items(root
);
10483 btrfs_btree_balance_dirty(root
);
10487 unlock_new_inode(inode
);
10492 /* Inspired by filemap_check_errors() */
10493 int btrfs_inode_check_errors(struct inode
*inode
)
10497 if (test_bit(AS_ENOSPC
, &inode
->i_mapping
->flags
) &&
10498 test_and_clear_bit(AS_ENOSPC
, &inode
->i_mapping
->flags
))
10500 if (test_bit(AS_EIO
, &inode
->i_mapping
->flags
) &&
10501 test_and_clear_bit(AS_EIO
, &inode
->i_mapping
->flags
))
10507 static const struct inode_operations btrfs_dir_inode_operations
= {
10508 .getattr
= btrfs_getattr
,
10509 .lookup
= btrfs_lookup
,
10510 .create
= btrfs_create
,
10511 .unlink
= btrfs_unlink
,
10512 .link
= btrfs_link
,
10513 .mkdir
= btrfs_mkdir
,
10514 .rmdir
= btrfs_rmdir
,
10515 .rename2
= btrfs_rename2
,
10516 .symlink
= btrfs_symlink
,
10517 .setattr
= btrfs_setattr
,
10518 .mknod
= btrfs_mknod
,
10519 .setxattr
= generic_setxattr
,
10520 .getxattr
= generic_getxattr
,
10521 .listxattr
= btrfs_listxattr
,
10522 .removexattr
= generic_removexattr
,
10523 .permission
= btrfs_permission
,
10524 .get_acl
= btrfs_get_acl
,
10525 .set_acl
= btrfs_set_acl
,
10526 .update_time
= btrfs_update_time
,
10527 .tmpfile
= btrfs_tmpfile
,
10529 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
10530 .lookup
= btrfs_lookup
,
10531 .permission
= btrfs_permission
,
10532 .get_acl
= btrfs_get_acl
,
10533 .set_acl
= btrfs_set_acl
,
10534 .update_time
= btrfs_update_time
,
10537 static const struct file_operations btrfs_dir_file_operations
= {
10538 .llseek
= generic_file_llseek
,
10539 .read
= generic_read_dir
,
10540 .iterate_shared
= btrfs_real_readdir
,
10541 .unlocked_ioctl
= btrfs_ioctl
,
10542 #ifdef CONFIG_COMPAT
10543 .compat_ioctl
= btrfs_compat_ioctl
,
10545 .release
= btrfs_release_file
,
10546 .fsync
= btrfs_sync_file
,
10549 static const struct extent_io_ops btrfs_extent_io_ops
= {
10550 .fill_delalloc
= run_delalloc_range
,
10551 .submit_bio_hook
= btrfs_submit_bio_hook
,
10552 .merge_bio_hook
= btrfs_merge_bio_hook
,
10553 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
10554 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
10555 .writepage_start_hook
= btrfs_writepage_start_hook
,
10556 .set_bit_hook
= btrfs_set_bit_hook
,
10557 .clear_bit_hook
= btrfs_clear_bit_hook
,
10558 .merge_extent_hook
= btrfs_merge_extent_hook
,
10559 .split_extent_hook
= btrfs_split_extent_hook
,
10563 * btrfs doesn't support the bmap operation because swapfiles
10564 * use bmap to make a mapping of extents in the file. They assume
10565 * these extents won't change over the life of the file and they
10566 * use the bmap result to do IO directly to the drive.
10568 * the btrfs bmap call would return logical addresses that aren't
10569 * suitable for IO and they also will change frequently as COW
10570 * operations happen. So, swapfile + btrfs == corruption.
10572 * For now we're avoiding this by dropping bmap.
10574 static const struct address_space_operations btrfs_aops
= {
10575 .readpage
= btrfs_readpage
,
10576 .writepage
= btrfs_writepage
,
10577 .writepages
= btrfs_writepages
,
10578 .readpages
= btrfs_readpages
,
10579 .direct_IO
= btrfs_direct_IO
,
10580 .invalidatepage
= btrfs_invalidatepage
,
10581 .releasepage
= btrfs_releasepage
,
10582 .set_page_dirty
= btrfs_set_page_dirty
,
10583 .error_remove_page
= generic_error_remove_page
,
10586 static const struct address_space_operations btrfs_symlink_aops
= {
10587 .readpage
= btrfs_readpage
,
10588 .writepage
= btrfs_writepage
,
10589 .invalidatepage
= btrfs_invalidatepage
,
10590 .releasepage
= btrfs_releasepage
,
10593 static const struct inode_operations btrfs_file_inode_operations
= {
10594 .getattr
= btrfs_getattr
,
10595 .setattr
= btrfs_setattr
,
10596 .setxattr
= generic_setxattr
,
10597 .getxattr
= generic_getxattr
,
10598 .listxattr
= btrfs_listxattr
,
10599 .removexattr
= generic_removexattr
,
10600 .permission
= btrfs_permission
,
10601 .fiemap
= btrfs_fiemap
,
10602 .get_acl
= btrfs_get_acl
,
10603 .set_acl
= btrfs_set_acl
,
10604 .update_time
= btrfs_update_time
,
10606 static const struct inode_operations btrfs_special_inode_operations
= {
10607 .getattr
= btrfs_getattr
,
10608 .setattr
= btrfs_setattr
,
10609 .permission
= btrfs_permission
,
10610 .setxattr
= generic_setxattr
,
10611 .getxattr
= generic_getxattr
,
10612 .listxattr
= btrfs_listxattr
,
10613 .removexattr
= generic_removexattr
,
10614 .get_acl
= btrfs_get_acl
,
10615 .set_acl
= btrfs_set_acl
,
10616 .update_time
= btrfs_update_time
,
10618 static const struct inode_operations btrfs_symlink_inode_operations
= {
10619 .readlink
= generic_readlink
,
10620 .get_link
= page_get_link
,
10621 .getattr
= btrfs_getattr
,
10622 .setattr
= btrfs_setattr
,
10623 .permission
= btrfs_permission
,
10624 .setxattr
= generic_setxattr
,
10625 .getxattr
= generic_getxattr
,
10626 .listxattr
= btrfs_listxattr
,
10627 .removexattr
= generic_removexattr
,
10628 .update_time
= btrfs_update_time
,
10631 const struct dentry_operations btrfs_dentry_operations
= {
10632 .d_delete
= btrfs_dentry_delete
,
10633 .d_release
= btrfs_dentry_release
,